Hi,

I have a "old" system that was coded in FPC 2.6.5. Today I had to change something in the code and now I need to update to FPC 3.0 and Lazarus 1.9.

This system uses a COM object. I made a class to wrap the configuration.

So, all string arguments in this class is WideString based. The SetLicence method will receive a WideString but the source is a "string".

Look:

    Lib.SetLicense(
      IniFile.ReadString('TheLib', 'license', '')
    );

As you know, IniFile.ReadString returns a "string" and some internal conversion is happening and the licence is not valid anymore.

If I put the licence as a string directly, it works:

    Lib.SetLicense(
      'my_licence_here'
    );

How can I change my code to work properly using Ini files, strings and WideString?

Best regards,
Marcos Douglas

On Sat, 12 Aug 2017 16:46:09 -0300
"Marcos Douglas B. Santos via Lazarus"

[...] Lib.SetLicense( IniFile.ReadString('TheLib', 'license', '') );

What encoding has the ini file?

Mattias

ANSI. A simple text file on Windows with only ANSI chars.

But I'm so sorry Mattias, it was my fault.
The program was reading the wrong file version (problem in paths...).

It works now, but I have one question:
What is the right way to code to do not see this warning?

Warning: Implicit string type conversion from "AnsiString" to "WideString"

Regards,
Marcos Douglas

On Sat, 12 Aug 2017 17:43:29 -0300
"Marcos Douglas B. Santos via Lazarus"

[...] > What encoding has the ini file? ANSI. A simple text file on Windows with only ANSI chars.

Which one? Do you mean Windows CP-1252?

[...] Warning: Implicit string type conversion from "AnsiString" to "WideString"

Explicit type cast:

Lib.SetLicense(
WideString(IniFile.ReadString('TheLib', 'license', '')) );

Mattias

Yes...
But would it make any difference?

Wow... everywhere? :(

Regards,
Marcos Douglas

"Marcos Douglas B. Santos wrote

> Which one? Do you mean Windows CP-1252? Yes... But would it make any difference?

I recently had a problem with an application that was converted from old string type to AnsiString and seemingly worked in the new Unicode environment.
However, we received reports that it had failed in some Asian countries (Korea, China, Thailand) and upon checking it turned out that the data inside a string used as buffer was changed because of locale differences....

After switching out the affected variable declarations from AnsiString to RawByteString the application seemingly started to work again also on these locations.

So AnsiString is not safe either....

And after this I have spent some time to totally rework the use of strings as buffers to instead use TBytes. Lots of work but guaranteed to not sneak in unexpected conversions.

Is not simpler to use RawByteString instead TBytes?

Regards,
Marcos Douglas

Well, initially just changing the declarations would seem to be simpler. But given how the conversion problem sneaked up behind my back, I thought it wiser to move all serial comm buffers from various string types (string->AnsiString->RawByteString) to TBytes since that is really guaranteed to be "the real thing".

Whenever there is a need for displaying the data or putting them into a string type variable I have added a few utility functions to do the conversions using the Move() procedure. Likewise I made a PosBin() for searching for patterns like Pos() for strings etc.

Bo Berglund wrote

So AnsiString is not safe either....

That is a little misleading.
Actually using the Windows system codepage is not safe any more. The current Unicode system in Lazarus maps AnsiString to use UTF-8. Text with Windows codepage must be converted explicitly. This is a breaking change compared to the old Unicode suppport in Lazarus 1.4.x + FPC 2.6.x.
The right solution is to use Unicode everywhere. Windows codepages can be seen as a historical remain, retained for backwards compatibility. Now is year 2017, Unicode has been used for decades. Everybody should use it by now.

Marcos Douglas, please change the encoding in your text file to UTF-8. Every decent text editor, including the editor in Lazarus, has a feature to do it.
Once the data is Unicode, it is all smooth sailing. Data is converted between UTF-8 and UTF-16 losslessly.

One more thing:
Data for WideString/UnicodeString parameters in WinAPI functions are converted automatically. You can ignore the warning or suppress it by a type cast as Mattias showed.
However for PWideChar parameters you should create an explicit temporary variable, usually UnicodeString but WideString for OLE. Assigning to it from your "String" data converts encoding. Then cast the new variable as the required pointer type.

Juha

Bo Berglund wrote

I recently had a problem with an application that was converted from old string type to AnsiString and seemingly worked in the new Unicode environment.

What was the old string type?

However, we received reports that it had failed in some Asian countries (Korea, China, Thailand) and upon checking it turned out that the data inside a string used as buffer was changed because of locale differences....

Unicode was designed to solve exactly the problems caused by locale differences. Why don't you use it?

RawByteString is for text which encoding is not meant to be converted. It has its special use cases.
TBytes is usually for binary data.
Did I understand right: you use TBytes to hold strings having Windows codepage encoding?
That sounds like a very dummy thing to do!
Again: Why not Unicode? Then you could throw away your hacks.

Juha

Note: The programs were started back in around 2000 using Delphi 7...

We used "string" as the container for processing serial data to/from CNC machine tool controllers amongst others. This was triggered really by the serial components, which mostly transferred char(acters) and had methods for sending and receiving strings, even though we usually used char.

Again, these are old existing programs and we are not doing this anymore for new programs. However, there is one problem still becauyse there is an interface point to the hardware, in the form of serial components, which still handle chars...
And chars are nowadays Unicode chars, i.e. not mapping to bytes sent by RS232...
And our data are NOT text, they are binary streams of bytes.

My first attempt at "fixing" the problem in Asian locales was to use RawByteString so as to inhibit conversions.
Still with these as comm buffers...
It seemed to work out, but to be safer I have reworked one application to replace with TBytes everywhere comm data are handled.

TBytes is usually for binary data.

Exactly, and this is why I made the comment that to be on the safe side dealing with RS232 the buffers should be TBytes (or some other similar construct).

Did I understand right: you use TBytes to hold strings having Windows codepage encoding?

No, definitively not. At the time we were not aware of any encoding at all. To us a string was just a handy container for the serial data like a dynamic array of byte with some useful functions available for searching and things like that. I think we were not alone...

Again: Why not Unicode? Then you could throw away your hacks.

The application itself is Unicode now but we had to run circles around the RS232 comm part. When converting to Unicode we first set the comm related strings to be AnsiString...

PS: We never programmed the serial interface directly, we always used commercial RS232 components and they all dealt with char and string... DS

Bo Berglund wrote

And our data are NOT text, they are binary streams of bytes.

I see. Then TBytes indeed is the best choice.
You have misused "String" or "AnsiString" from the beginning for binary data. There have always been warnings against it.
The new Lazarus Unicode system did not create the problem but made it more visible.

Marcos Douglas however has a different problem.
Your recommendation to use RawByteString or TBytes does not apply in his case and thus was a bit misleading.

Juha

Juha Manninen wrote

You have misused "String" or "AnsiString" from the beginning for binary data. There have always been warnings against it.

While this might be true, it's decently silly, IMHO.

The name "String" can easily be interpreted as "String of things" and does not necessarily mean "String of printable stuff".

The management Pascal always provided for strings (after the "Short String" was not the only string type) (i.e. Operators, built-in functions, lazy copy, reference counting) is perfectly applicable to "Strings of things", and don't force any known drama only was introduced by Embarcadero's abysmal / sloppy implementation of automatic code conversion for strings.

-Michael

Juha Manninen wrote

Unicode was designed to solve exactly the problems caused by locale differences. Why don't you use it?

I believe you effectively answer your own question in your preceding post:

If you are processing strings as "text" then you probably do not care how it is encoded and can live with "breaking changes". However, if, for some reason you are or need to be aware of how the text is encoded - or are using string types as a useful container for binary data then, types that sneak up on you with implicit type conversions or which have semantics that change between compilers or versions, are just another source of bugs.

PChar used to be a safe means to access binary data - but not anymore, especially if you move between FPC and Delphi. (One of my gripes is that the FCL still makes too much use of PChar instead of PByte with the resulting Delphi incompatibility). The "string" type also used to be a safe container for any sort of binary data, but when its definition can change between compilers and versions, it is now something to be avoided.

As a general rule, I now always use PByte for any sort of string that is binary, untyped or encoding to be determined. It works across compilers (FPC and Delphi) with consistent semantics and is safe for such use.

I also really like AnsiString from FCP 3.0 onwards. By making the encoding a dynamic attribute of the type, it means that I know what is in the container and can keep control.

I am sorry, but I would only even consider using Unicodestrings as a type (or the default string type) when I am just processing text for which the encoding is a don't care, such as a window caption, or for intensive text analysis. If I am reading/writing text from a file or database where the encoding is often implicit and may vary from the Unicode standard then my preference is for AnsiString. I can then read the text (e.g. from the file) into a (RawByteString) buffer, set the encoding and then process it safely while often avoiding the overhead from any transliteration. PByte comes into its own when the file contains a mixture of binary data and text.

Text files and databases tend to use UTF-8 or are encoded using legacy Windows Code pages. The Chinese also have GB18030. With a database, the encoding is usually known and AnsiString is a good way to read/write data and to convey the encoding, especially as databases usually use a variable length multi-byte encoding natively and not UTF-16/Unicode. With files, the text encoding is usually implicit and AnsiString is ideal for this as it lets you read in the text and then assign the (implicit) encoding to the string, or ensure the correct encoding when writing.

And anyway, I do most of my work in Linux, so why would I even want to bother myself with arrays of widechars when the default environment is UTF8?

We do need some stability and consistency in strings which, as someone else noted have been confused by Embarcadero. I would like to see that focused on AnsiString with UnicodeString being only for specialist use on Windows or when intensive text analysis makes a two byte encoding more efficient than a variable length multi-byte encoding.

Tony Whyman
MWA

"The right solution is to use Unicode everywhere." I agree. But would be best if the compiler uses Unicode everywhere and us, developers, using just one type called "string"... Even if this break the old code. Maybe, instead using "string", the new code should be use just UnicodeString...

Well, I know that many people here already had this "fight" about Unicode so, let's forget about it what the compiler "should" or not to do.

You're right.

This is a ugly trick... but I understood what you mean.

Best regards,
Marcos Douglas

Unicode everywhere and you using AnsiString and doing everything... Now I'm confused.

And anyway, I do most of my work in Linux, so why would I even want to bother myself with arrays of widechars when the default environment is UTF8?

Maybe you do not have problems because you don't use Windows.

FPC and Lazarus claim they are cross-platform — this is a fact — and because that, IMHO, both should be use in only one way in every system, don't you think?

Best regards,
Marcos Douglas

Marcos Douglas B. Santos wrote

"The right solution is to use Unicode everywhere."

Embarcadero though that this would not b the "right" solution. Otherwise they would not have invented the encoding aware strings.

IMHO that was a good idea. They only completely failed to do a decent specification and implementation.

-Michael

Precisely. But why this fixation on UTF-16/Unicode and not UTF8?

Lazarus is already a UTF8 environment.

Much of the LCL assumes UTF8.

UTF8 is arguably a much more efficient way to store and transfer data

UTF-16/Unicode can only store 65,536 characters while the Unicode standard (that covers UTF8 as well) defines 136,755 characters.

UTF-16/Unicode's main advantage seems to be for rapid indexing of large strings.

You made need UTF-16/Unicode support for accessing Microsoft APIs but apart from that, why is it being promoted as the universal standard?

[...] Lazarus is already a UTF8 environment. Much of the LCL assumes UTF8.

True.

UTF8 is arguably a much more efficient way to store and transfer data

It depends.

UTF-16/Unicode can only store 65,536 characters while the Unicode standard (that covers UTF8 as well) defines 136,755 characters.

No.
UTF-16 can encode the full 1 million Unicode range. It uses one or two words per codepoint. UTF-8 uses 1 to 4 bytes.
See here for more details:
https://en.wikipedia.org/wiki/UTF-16

Although you are right, that there are still many applications, that falsely claim to support UTF-16, but only support the first $D800 codepoints.

UTF-16/Unicode's main advantage seems to be for rapid indexing of large strings.

That's only true for UCS-2, which is obsolete.

You made need UTF-16/Unicode support for accessing Microsoft APIs but apart from that, why is it being promoted as the universal standard?

Who does that?

Mattias

Juha Manninen wrote

Now is year 2017, Unicode has been used for decades. Everybody should use it by now.

Indeed, I can't agree more. Plus, I normally use UTF-8 for any text files I create.

Regards,
Graeme

Because the obvious implication when someone argues against AnsiString (from which UTF8String derives) and talks about Unicode is that they are promoting UTF-16 and the UnicodeString type. Perhaps this is because I am old enough to remember when MS first added wide characters to Windows and that they called it "Unicode". To me, when people say "Unicode" they mean Windows wide characters.

Perhaps the problem is the use of the word "Unicode". By trying to embrace UTF8, UTF16 and UTF32 with the older UCS-2 it is perhaps too ambiguous a term - especially as the Delphi/FPC UnicodeString type exists and probably (but I'm not certain) means UTF-16.

What I see in FPC/Lazarus today is:

- UTF8 supported through AnsiString.

- A confusion of Widestring/UnicodeString for UTF-16 and legacy UCS-2.

- Nothing for UTF-32.

If nothing else, FPC Lazarus could do with a clean-up of both terminology and string types. Indeed, why isn't there a single container string type for all character sets where the encoding whether a legacy code page, UTF8, UTF16 or UTF32 is simply a dynamic attribute of the type - a sort of extended AnsiString?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170814/38f36890/attachment.html>

Tony Whyman wrote

ambiguous a term - especially as the Delphi/FPC UnicodeString type exists and probably (but I'm not certain) means UTF-16.

Yes, that is f**ken annoying. FPC should have named it what it really is
- UTF16String! But instead they followed Delphi like a lemming and named it UnicodeString.

In reality, UNICODE means text with an encoding of any of UTF-8, UTF-16LE, UTF-16BE, or UTF-32.

In terms of Delphi and FPC, they decided Unicode = UTF-16. I'm not even sure if they mean LE or BE.

Regards,
Graeme

Am 14.08.2017 16:11 schrieb "Tony Whyman via Lazarus" <

: If nothing else, FPC Lazarus could do with a clean-up of both terminology

and string types. Indeed, why isn't there a single container string type for all character sets where the encoding whether a legacy code page, UTF8, UTF16 or UTF32 is simply a dynamic attribute of the type - a sort of extended AnsiString?

The main problem of such a dynamic type would be the inability to do fast indexing as the compiler would need to insert runtime checks for the size of a character. I had already thought the same, but then had to discard the idea due to this.

Regards,
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170814/6e5e7011/attachment.html>

Am 14.08.2017 16:21 schrieb "Graeme Geldenhuys via Lazarus" <

- UTF16String! But instead they followed Delphi like a lemming and named it UnicodeString.

Because the crowd demanding Delphi compatibility is larger than the crowd demanding exact terminology.

In reality, UNICODE means text with an encoding of any of UTF-8,

UTF-16LE, UTF-16BE, or UTF-32.

In terms of Delphi and FPC, they decided Unicode = UTF-16. I'm not even

sure if they mean LE or BE.

If I remember correctly it depends on the endianess of the platform... Though I could be wrong.

Regards,
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170814/d03dcac5/attachment.html>

I have no fixation in any Unicode flavors...
My "problem" is because I use Windows, not Linux where UTF8 is the default.

I didn't propose that.
But take a look in other languages, see what they are using.

Best regards,
Marcos Douglas

Tony Whyman wrote

Indeed, why isn't there a single container string type for all character sets where the encoding whether a legacy code page, UTF8, UTF16 or UTF32 is simply a dynamic attribute of the type - a sort of extended AnsiString?

As Sven Barth wrote, they have different size of char.

Tony Whyman, this issue has been discussed again and again for the past 10+ years first in FPC mailing lists and then in Lazarus lists. The current Unicode support in Lazarus works f***ing well and is amazingly compatible with Delphi.
WinAPI parameters may require an explicit temporary UnicodeString variable but even then the code is compatible with Delphi.

Tony Whyman, Marcos Douglas and Michael Schnell, please study the facts. For starters, this is about the current Unicode support in Lazarus: http://wiki.freepascal.org/Unicode_Support_in_Lazarus I think the dynamic encoding and automatic conversion now work perfectly well. If you have a piece of code where it does not work, please ask for detailed info.

Juha

Sven Barth wrote

Because the crowd demanding Delphi compatibility is larger than the crowd demanding exact terminology.

... or even a revised concept avoiding the junk presented by Embarcadero :(

But obviously the fpc team has no choice.

-Michael

What "indexing" do you think of ?
Could you give an example where such a difference is supposed to get important ?

(As you know I wrote a paper where I claimed the contrary. I'd like to revise same if necessary.)

-Michael

Sven Barth wrote

The main problem of such a dynamic type would be the inability to do fast indexing as the compiler would need to insert runtime checks for the size of a character. I had already thought the same, but then had to discard the idea due to this.

Is this really a big problem? It is not as if it would be necessary to do a table lookup everytime you index a string as the indexing method could be an attribute of the string and updated with the character encoding attribute. Is it really that complicated for the compiler to generate code that jumps to an indexing method depending upon a data attribute?

Is your problem really more about the result type as, depending on the character width, the result could be an AnsiChar or WideChar or a UTF8 character for which I don't believe there is a defined char type (other than an arguable mis-use of UCS4Char)?

I can accept that a clear up of this area would also have to extend to the char types as well - but I would also argue that that is well overdue. On a quick count, I found 7 different char types in the system unit.

On Sat, 12 Aug 2017 17:56:58 -0300
"Marcos Douglas B. Santos via Lazarus"

Just

You could instead define an overloaded Lib.SetLicense(AnsiString). Or you could disable this hint altogether for your project (not recommended). Select the message in the Messages window. Right click and click on add -vm....

Mattias

IMHO the main problem of adding a new string type is https://xkcd.com/927/

Mattias

If a topic keeps on being discussed after 10+ years of argument, the reason is usually either (a) the problem and its solution have not been documented properly, or (b) the outcome is an unsatisfactory compromise.

In this case, I would argue that both are true.

I went back and read the wiki article you mentioned and was no more the wiser as to why the current mess exists. Is it really no more than because Delphi continues to screw up in this area, so must FPC? The body of the article appears to be a set of notes - not necessarily wrong in themselves but lacking the background and context needed to explain why it is like it is.

This problem will keep coming up until it is fixed properly and, by that, I mean the that solution is consistent, understandable intuitively and well documented. Windows eccentricity also need to kept to Windows.

Here is my wish list:

1. Stop using the term "Unicode".

It is too ambiguous. It is used as both an all embracing term for multi-byte encoding and as a synonym for UTF16 and that is really too confusing. The problem is made worse by having UnicodeString as a two byte wide string type in both FPC and Delphi.

2. Clean up the char type.

When Wirth created the "char" type in Pascal it was a simple ASCII or EBCDIC character. There are now seven different char types (including type equivalence) with no guidelines on when each is applicable. This is too many. Why shouldn't there be a single char type that intuitively represents a single character regardless of how many bytes are used to represent it. Yes, in a world where we have to live with UTF8, UTF16, UTF32, legacy code pages and Chinese variations on UTF8, that means that dynamic attributes have to be included in the type. But isn't that the only way to have consistent and intuitive character handling?

3. The problem with string handling today is that it is not based on a consistent approach to the character type.

If you clean up character handling then the model for string handling should become obvious. A string is after all no more than a container for a character array and which should be constrained to have the same character encoding. A string should intuitively represent a string of text regardless of how many bytes are used to represent each character and with dynamic attributes to tell you how it is encoded.

4. FPC should clean up Delphi's mess for it. If a unified string type follows a consistent model then it should be possible to make all Delphi string types synonyms.

You will need to allow exceptions for legacy programs that insist on manipulating the bytes themselves - but that is not rocket science. There is also the issue of the Windows API and its insistence on Wide Strings - but isn't that why calling conventions such as cdecl and stdcall exist - to tell the compiler when it needs to reformat the call for a given API convention.

Tony Whyman

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170815/93b53d87/attachment-0001.html>

You can me as a "like" on that one.

Mattias Gaertner wrote

IMHO the main problem of adding a new string type is https://xkcd.com/927/

Exactly. I don't think we should add even more.

As it is now, FPC offers a way out for all cases:

WideString/UnicodeString for those that want 2-byte characters. A codepage-aware single-byte string for those that want 1-byte characters. The shortstring is even still available.

Attempting to store binary data in a string is not advisable. Dynamic arrays, TBytes and - in the worst case - TBytesStream are powerful enough to cover most use-cases in this area.

Michael.

Michael Van Canneyt wrote

WideString/UnicodeString for those that want 2-byte characters. A codepage-aware single-byte string for those that want 1-byte characters. The shortstring is even still available.

IM (often stated) O, this does not help as long as TStrings does not without forced auto-conversion support the string type the user is inclined to choose.

This obviously requires an (additional) fully dynamic string brand.

This (again obviously) is not the "Embarcadero way", but supposedly does not necessarily lead to incompatibility regarding the user code.

-Michael

Please check TStrings in trunk. This exists.

procedure LoadFromFile(const FileName: string; AEncoding: TEncoding); overload; virtual;
procedure LoadFromStream(Stream: TStream; AEncoding: TEncoding); overload; virtual;

The only 'problem' is that TStrings uses a single-byte string.

This cannot be solved properly except by duplicating the classes unit.

Michael.

Tony Whyman wrote

In this case, I would argue that both are true.

And the culprit obviously is Embarcadeo and not the fpc or the Lazarus team, who did their best to try to do a compatible and implementation that is really workable on the multiple supported platforms (which E$ did not feel necessary when they released the encoding aware strings).

Maybe a better solution can be found, but who would want to nudge the fpc / Lazarus developers to invest a huge amount of time to create it and then make sure it is decently tested stable ?

-Michael

Michael Van Canneyt wrote

This cannot be solved properly except by duplicating the classes unit.

Sorry to disagree, but IMHO this can only be solved properly by defining an additional fully dynamically encoded string type and use same for TStrings (see ->
http://wiki.freepascal.org/not_Delphi_compatible_enhancement_for_Unicode_Support )

But I am perfectly aware that implementing this would be a huge effort (see other mail here), and nobody i entitled to ask for this. (I wrote the article just to elaborate what was discussed in the fpc mailing list at that time.)

-Michael

It does not explain what the characters of DynamicString are, does it?

Mattias

I was just going to write that.

The problem of the element size is circumvented by simply not digging into it.

What does S[2] mean in your proposal ? Is it 1, 2, 4 or even 8 bytes ?

Michael.

Mattias Gaertner wrote

It does not explain what the characters of DynamicString are, does it?

I don't understand what you are asking.

The element size and encoding of a Dynamic String ("CP_ANY" in the document) are not predefined, but depend on the content:

http://wiki.freepascal.org/not_Delphi_compatible_enhancement_for_Unicode_Support -> Defining String variables and String types:

*CP_ANY* = $FF00 // ElementSize dynamically assigned // fully dynamical String for intermediate storing string content // just assigned to the Type or variable, never used in the "Encoding" field in the string header.

Hence it stores the "branding" when it is assigned to from a string with a fixed branding (such as *CP_UTF8*), and the content is auto-converted if necessary when assigning form CP_ANY to a fixed branded string variable.

If (in your example) the data is read from a file, a CP_ANY Strings based StringList would keep the encoding/char_size of the data as t is in the file (it would need to somehow get to know the presumed encoding of the file, anyway) and store that information in the EncodingBrandNumber and ElementSize fields (which do exist in any "NewString" variable, anyway), in each String read.

If the user assignes an element of the stringlist to a fixed branding (such as *CP_UTF8*), the content obviously is auto-converted if necessary when assigning form CP_ANY to a fixed branded string variable, as usual.

In fact I suppose that the current implementation of TStringlist does not use new strings to store the data on the heap, but I never said that trying to implement such idea would not require a lot of work.

-Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170815/31b153f1/attachment.html>

Michael Van Canneyt wrote

What does S[2] mean in your proposal ? Is it 1, 2, 4 or even 8 bytes ?

Regarding the users' appreciation, the S[x] notation is decently incompatible between the different string types and compiler versions.

There were hundreds of complains in all the appropriate forums and mailing list.

So not much additional harm can be done, anyway.

I suggest that it should be according to the character_size definition stored S, and the operation c := S[x] should transfer the appropriate count of bits, provided the type of c allows for taking them.

This seems to be compatible to the current implementation of any 1-Byte brand and UTF16.

-Michael

Michael Schnell wrote

On 15.08.2017 12:15, Michael Van Canneyt via Lazarus wrote: > What does S[2] mean in your proposal ? Is it 1, 2, 4 or even 8 bytes ? Regarding the users' appreciation, the S[x] notation is decently incompatible between the different string types and compiler versions.

Of course not.

It's 1 byte for ansistring, 2 bytes for widestring.

The point is that the compiler knows how many bytes it is based on the declaration of S. In your proposal, it is dynamic, if I understand it correctly.

There were hundreds of complains in all the appropriate forums and mailing list.

Complaints about what exactly ?

As far as I understand your proposal, this currently cannot be done ?

The compiler needs to know the S[X] size at compile time.

Michael.

You would have to define what "a single character" means in the first place. This is especially important when it involves precomposed characters and combining characters.

Bart

Tony Whyman wrote

Why shouldn't there be a single char type that intuitively represents a single character regardless of how many bytes are used to represent it.

I suppose by "char" you mean "single printable thingy" with Unicode it's rather debatable what such a thingy is.

Hence a Unicode singe char would need to be just be a Unicode string.

-Michael

see ->
http://wiki.freepascal.org/not_Delphi_compatible_enhancement_for_Unicode_Support

-Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170815/a2e34a24/attachment.html>

Do you mean a 'char' is a string in your proposal?

Mattias

That would be a neat recursive definition :)

Michael.

Mattias Gaertner wrote

Do you mean a 'char' is a string in your proposal?

Nope. In my proposal there would be Chars for any statically encoded String Type, hence 1, 2, 4, and 8 byte wide. (As regarding statically encoded string (and char) brands, it's just an extension of the existing paradigm.

I did not think about the necessity to also have a dynamically encoded Char type. If yes, it (like a string) would need the additional fields for encoding number and bytes_per_char, and the appropriate compiler magic to handle them appropriately (workalike to a on-element string).

-Michael

8 bytes?

Do you propose a string without the array operator [] ?

Mattias

Michael Van Canneyt wrote

The only 'problem' is that TStrings uses a single-byte string.

Why can't that be changed to a UnicodeString or UTF8String - after all, the Unicode standard is meant to support all languages. I would have thought that would be an obvious move for a Unicode-aware RTL. TStrings could also be extended (if it hasn't already) to keep track of what encoding is read in from file, and what encoding in should procedure when lines are extracted - in case those two encodings are not the same.

Regards,
Graeme

El 15/08/17 a les 11:25, Michael Van Canneyt via Lazarus ha escrit:

I has worked extremely well and reliably until fpc 2.6.4 (i.e. with string=ansistring).
Does it not work in 3.x?
If not it's a big problem, not only for my code (that I can, reluctantly, change) but for 3rd party libraries/components (e.g. synapse comes to mind)

Bye

Michael Van Canneyt wrote

As it is now, FPC offers a way out for all cases: WideString/UnicodeString for those that want 2-byte characters.

what if 3 and 4 byte characters are required? will they also work in UnicodeStrings?

i'm looking at this from a linux POV but have been trying to come from the very old school DOS TP stuff using codepages... especially needing to be able to read codepage strings and properly convert all their characters to UTF-8...

converting back would be a huge help, too... even with the possible loss of characters requiring replacing them with "?" or something to hold their place and show they didn't convert... that or even leaving them in their 2, 3 or 4 byte form and let those using older codepage stuff see them raw...

Luca Olivetti wrote

but for 3rd party libraries/components (e.g. synapse comes to mind

Then better start filing bug reports to all those 3rd party libraries and components - they have been abusing the system and will silently fail. Not to mention that FPC is and the new string changes were introduced in v3.0.0 already.

Regards,
Graeme

El 15/08/17 a les 21:14, Graeme Geldenhuys via Lazarus ha escrit:

Wait a minute, why "abuse"?
After all, before code aware strings, an ansistring could store any kind of arbitrary data with no problem and no conversion, and made it extremely easy to, e.g., add bytes to a buffer or find and extract data from the same buffer.
*If* code that worked before (and dare I say without abusing the language) suddenly breaks, the bug is in the compiler and not in the library.
(I remarked the "if" because I don't know if that's the case, according to Bo Berglund's experience it is)

Bye

[...] *If* code that worked before (and dare I say without abusing the language) suddenly breaks, the bug is in the compiler and not in the library.

... unless of course the incompatibility is deliberate and documented. In this case it is.

Mattias

Furthermore, if you use(d) strings for binary data, just replace old string for AnsiString/RawByteString (and Char for AnsiChar, PChar for PAnsiChar) and you are good to go. Annoying but no big deal.

Ondrej

El 15/08/17 a les 21:38, Ondrej Pokorny via Lazarus ha escrit:

If that's all it's OK then, thank you.

Bye

El 15/08/17 a les 21:38, Ondrej Pokorny via Lazarus ha escrit:

If that's all it's OK then, thank you.

Bye

El 15/08/17 a les 22:08, Luca Olivetti ha escrit:

Sorry for the direct reply, it was meant for the list.

Bye

Luca Olivetti wrote

Wait a minute, why "abuse"? After all, before code aware strings, an ansistring could store any kind of arbitrary data with no problem and no conversion, and made it extremely easy

Just listen to what you are saying.... A string type and you want to store all kinds of non-string related data in that type. How is that not "abuse"??? Use a TBytes, TStream or other binary byte based storage mechanism. A string type was definitely not the right choice.

Regards,
Graeme

El 15/08/17 a les 22:45, Graeme Geldenhuys via Lazarus ha escrit:

A "string" was just a handy container for bytes so I think it was the right choice for storing, er, bytes.

Bye

Luca Olivetti wrote

A "string" was just a handy container for bytes so I think it was the right choice for storing, er, bytes.

The type "String" has always been an alias to another type, and could mean many things. eg: ShortString, AnsiString, and now UnicodeString. Making the assumption that it will always be a container for byte sized data was wrong.

In hind sight, using TBytes or TMemoryStream and it would have been very clear that it is a storage container for byte sized data, and no automatic conversion (by the compiler) would be done to data stored in such containers.

Don't worry though, you were not alone in making that wrong assumption. Many Delphi developers have made that mistake, and some are still making that mistake today.

Regards,
Graeme

Luca Olivetti wrote

(I remarked the "if" because I don't know if that's the case, according to Bo Berglund's experience it is)

Just to expand on my "experience" and the reason I posted:

My work on converting the old program started back a couple of years when I went from Delphi 2007 (pre-unicode) to Delphi XE5 because we wanted the GUI to be translatable to non-western languages.

But then all the communications functions (and these are many in this utility application) broke because they used strings as containers for the inherently binary serial data.

So I followed advice on the Embarcadero forum to switch to AnsiString because that was really what the old string type was an alias for. I had no great insight in the inner workings of the string handling functions but I "knew" that AnsiString was a 1-byte per element and (unicode)string was now a 2-byte per element container. The fact that the code could alter the content of the AnsiString did not dawn on me at all.
And the comm functions worked fine after the change (I tested a lot, but of course only on my English Win7 computer).

Then some time ago there was a report of a failure of the new program version that only happened in Korea, China and Thailand. In the log files there was a very strange entry about finding an illegal command byte when sending a command to the equipment.

It never triggered when I debugged the problem, for me and my collegues it worked flawlessly. So I had to add more logging and found that the problem arose when the outgoing command was built. A certain 1-byte command was then expanded to 2 bytes with the wrong first byte! The commands in the protocol are the first byte of the data of a telegram and they are in range $C0..$E9.
When one of these (I don't now remember exactly which one) was used in an assignment to the AnsiString buffer it was converted to $3F + something that was never logged and the operation failed because the equipment could not decode the command.

So I asked again on the forum and was steered towards RawByteString because presumably that container would disallow conversions. And when I changed this and sent a new version to the distributor in Korea the problem was seemingly gone.

Based on this experience I wanted to alert the OP of the fact that using AnsiString instead of string is not a cure-all for binary data, you need to fix the codepage too, which is what the RawByteString does for you....

But I have now moved on and replaced all comm related containers with TBytes including modifying the serial component we have used. (With some help from Remy Lebeau).

Bo Berglund wrote

But I have now moved on and replaced all comm related containers with TBytes including modifying the serial component we have used. (With some help from Remy Lebeau).

I forgot to mention that the problem area is located inside a non-GUI class file for handling the communications, and this file is also used in some programs written in FPC for Raspberry Pi target computers. I.e. Linux and the reason for going to FPC.
So I want it to be both FPC and Delphi compatible...

Graeme Geldenhuys wrote

Why can't that be changed to a UnicodeString or UTF8String

IMHO, any implementation of TStrings that forces a conversion (just because the class uses TStrings and not due to a logical demand), is a contradiction to providing code aware

Luca Olivetti wrote

I has worked extremely well and reliably until fpc 2.6.4 (i.e. with string=ansistring). Does it not work in 3.x?

I understand that storing uncoded Bytes in UTF8-Strings (hence in fpc) works as good as it always had, as long as all strings are defined with the same code branding as TSrings (and friends) is (i.e. UTF8), because there never will be a conversion.

But it does not work in Delphi, as here TStrings is defined to be UTF-16.

-Michael

Ondrej Pokorny wrote

Furthermore, if you use(d) strings for binary data, just replace old string for AnsiString/RawByteString (and Char for AnsiChar, PChar for PAnsiChar) and you are good to go. Annoying but no big deal.

This only works if all tools that you use do the same. And a major tool for handling strings is TStrings and it's siblings. You hardly an avoid using same.

-Michael

This thread is going out of topic.
Please start a new thread if you want to discuss Delphi strings.

Mattias

Bo Berglund wrote

Based on this experience I wanted to alert the OP of the fact that using AnsiString instead of string is not a cure-all for binary data, you need to fix the codepage too, which is what the RawByteString does for you....

Bo, everybody has known for decades that AnsiString is not for binary data. Why do you proclaim it as a new discovery?
The OP's problem was completely different. It was about text encoding. TBytes is clearly the right choice for your binary data, but this discussion is not about binary data!

What means "AnsiString instead of string"?
String is typically an alias for AnsiString.

Your sentence about RawByteString is also wrong.
There is no automatic codepage conversion for RawByteString.

Juha

Graeme Geldenhuys wrote

How is that not "abuse"???

IMHO it's a major shortcoming to define "string" as "printable text". In fact the name "String" does not suggest "string" in my understanding just is a sequence of similar "things".

A string type was definitely not the right choice.

Notwithstanding the discussion about the mere wording, this only would hold, if the system would provide a differently named non "printable text" basic type that comes with the features needed for such usage: reference counting, lazy copy, simple operators for concatenating and element extraction and replacement, built-in function for substring locating, ...

-Michael

Michael Schnell wrote

On 15.08.2017 22:45, Graeme Geldenhuys via Lazarus wrote: > How is that not "abuse"??? IMHO it's a major shortcoming to define "string" as "printable text".

On the contrary. That is exactly what it means.
Anything else is just a collection of bytes.

Michael.

Michael Schnell wrote

IMHO, any implementation of TStrings that forces a conversion (just because the class uses TStrings and not due to a logical demand), is a contradiction to providing code aware in FPC 3.x (using modern compiler modes - not TP or Mac) String =

UnicodeString. So it makes sense that TStrings should use UnicodeString internally to store its data. The Unicode standard is also the only standard that can support any language. So all Windows code-pages can be supported with the single UnicodeString type.

Are you suggesting that internally TStrings should have different storage for all possible languages, or some RawByteString type? So if you load some non-Latin code-page text internally it still stores that text as that non-Latin bytes? That would just over-complicate the TStrings class. FPC is moving towards UnicodeString being used internally for everything in the RTL, so why must TStrings be any different.

Regards,
Graeme

Mattias Gaertner wrote

This thread is going out of topic. Please start a new thread if you want to discuss Delphi strings.

You can't discuss fpc's string problems without mentioning Delphi, as they are a direct consequence as well of Delphi-compatibility as of Delphi-incompatibility.

-Michael

Marcos Douglas B. Santos wrote

Unicode everywhere and you using AnsiString and doing everything... Now I'm confused.

Yes, please read:
http://wiki.freepascal.org/Unicode_Support_in_Lazarus I have advertised it so much that some people are already irritated, but maybe you missed it so far.

FPC and Lazarus claim they are cross-platform — this is a fact — and because that, IMHO, both should be use in only one way in every system, don't you think?

Yes, and that's how it works.

This is a ugly trick... but I understood what you mean.

This was about the explicit temporary UnicodeString variable for WinAPI call parameters.
No, it is not ugly, the code remains 100% compatible with Delphi. Please remember also that direct WinAPI call are not needed in cross-platform code.

Juha

wkitty42--- wrote

what if 3 and 4 byte characters are required? will they also work in UnicodeStrings?

UTF-8 and UTF-16 are just encoding variants for 32 bit Unicode "characters", storing them in n (or 2*n) Bytes according to a simple scheme.

-Michael

Mattias Gaertner wrote

Do you propose a string without the array operator [] ?

I don't understand what you mean by this.

Of course an appropriate "char" type for each string encoding brand could to be provided, hence a "CP_QWord Char" as an alias or a QWord.

(Please keep in mind that in that paper (as explicitly pointed out) "String" is not a synonym for "printable text" but for "sequence of similar things". And here of course (at least in a 64 bit system) it's extremely appropriate to allow for 64 bit elements. And of course this is just a suggestion that could solve a certain class of problems but needs a big effort to do and verify the modifications in the compiler and the libraries.)

-Michael

The original post was about a string conversion warning.

Anyone who wants to discuss the grand picture of strings in FPC for the millionth time should start a new topic.

Mattias

Graeme Geldenhuys wrote

So it makes sense that TStrings should use UnicodeString internally to store its data. The Unicode standard is also the only standard that can support any language.

But in fact "Unicode" is just a universal standard defining 64 bit entities. The encoding of those varies: UTF-8, UTF-16 high byte first, UTF-16 low byte first, 64 bit low byte first, 64 bit high byte first, .... fpc and Delphi do support several of those as a string encoding (and with that crating any number of problems).

-Michael

Graeme Geldenhuys wrote

Are you suggesting that internally TStrings should have different storage for all possible languages,

the said paper I point out that a new fully dynamical string encoding brand would be introduced and same is used for TStrings. Everything else will not provide an improvement of the class of problems under discussion since years.

-Michael (knowing that this will never happen)

Mattias Gaertner wrote

Anyone who wants to discuss the grand picture of strings in FPC for the millionth time should start a new topic.

Right you are. And it will be by far too late and futile, anyway, because of the reasons discussed a million times.

-Michael

[...] But in fact "Unicode" is just a universal standard defining 64 bit entities.

No.
1,114,112 possible code points need at most 21 bits. Due to encoding at most 32bit.

Mattias

Tony Whyman wrote

UTF-16/Unicode can only store 65,536 characters while the Unicode standard (that covers UTF8 as well) defines 136,755 characters. UTF-16/Unicode's main advantage seems to be for rapid indexing of large strings.

That shows complete ignorance from your side about Unicode. You consider UTF-16 as a fixed-width encoding. :( Unfortunately many other programmers had the same wrong idea or they were just lazy. The result anyway is a lot of broken UTF-16 code out there.

Tony Whyman wrote

If a topic keeps on being discussed after 10+ years of argument, the reason is usually either (a) the problem and its solution have not been documented properly, or (b) the outcome is an unsatisfactory compromise.

Or (c) The people discussing are ignorant about the topic.

Hmmm...
Originally the page was a mess because it had lots of irrelevant background info about the old obsolete LCL Unicode support. Text was added by many people but none was removed.
Finally I cleaned the page. It now has most relevant info at the top and then special cases and technical details later. I am rather happy with the page now, it explains how to use Unicode with Lazarus as clearly as possible.
However I am willing to improve it. What kind of background and context would you need?

1. Stop using the term "Unicode".

You can stop using it. No problem.
For others however it is a well defined international standard. See: https://en.wikipedia.org/wiki/Unicode

2. Clean up the char type. ... Why shouldn't there be a single char type that intuitively represents a single character regardless of how many bytes are used to represent it.

What do you mean by "a single character"?
A "character" in Unicode can mean about 7 different things. Which one is your pick?
This question is for everybody in this thread who used the word "character".

Yes, in a world where we have to live with UTF8, UTF16, UTF32, legacy code pages and Chinese variations on UTF8, that means that dynamic attributes have to be included in the type. But isn't that the only way to have consistent and intuitive character handling?

What do you mean? Chinese don't have a variation of UTF8. UTF8 is global unambiguous encoding standard, part of Unicode.

The fundamental problem is that you want to hide the complexity of Unicode by some magic String type of a compiler.
It is not possible. Unicode remains complex but the complexity is NOT in encodings!
No, a codepoint's encoding is the easy part. For example I was easily able to create a unit to support encoding agnostic code. See unit LazUnicode in package LazUtils.
The complexity is elsewhere:
- "Character" composed of codepoints in precomposed and decomposed (normalized) forms.
- Compare and sort text based on locale.
- Uppercase / Lowercase rules based on locale.
- Glyphs
- Graphemes
- etc.

I must admit I don't understand well those complex parts. I do understand codeunits and codepoints, and I understand they are the easy part.

Juha

Mattias Gaertner wrote

1,114,112 possible code points need at most 21 bits. Due to encoding at most 32bit.

Sorry. Typo.
-Michael

Michael Schnell wrote

UTF-8 and UTF-16 are just encoding variants for 32 bit Unicode "characters", storing them in n (or 2*n) Bytes according to a simple scheme.

No, they are encodings for codepoints, not "characters" (whatever that means).

Michael Schnell, your posts are completely out of topic. Unicode related topics clearly pull you like a magnet and then you loose all control and start to proclaim your grand plan for a string revamp.
It can continue for months as we remember from past years. You should stop writing in this thread now. I agree with Mattias.

Juha

Juha Manninen wrote

You should stop writing in this thread now. I agree with Mattias.

I perfectly agree with you. But you can't blame me for answering when asked.

-Michael

Juha Manninen wrote

Unfortunately many other programmers had the same wrong idea or they were just lazy. The result anyway is a lot of broken UTF-16 code out there.

Yeah, I see that even in commercial products and projects. It's very sad to see. Hence I always promote UTF-8, and you can't get it wrong as easily as UTF-16. No endianess to worry about, no surrogate pairs and UTF-8 is ready for streaming (network or disk) out of the box.

Regards,
Graeme

Thanks. I know about this page... unfortunately looks like it is not enough, since many others still complain.

This thread is not only about WinAPI. I have this problem because I need to use a Windows 3rd Lib, which uses WideString.

Best regards,
Marcos Douglas

Marcos Douglas B. Santos wrote

Thanks. I know about this page... unfortunately looks like it is not enough, since many others still complain.

What is missing? I can try to improve it.

This thread is not only about WinAPI. I have this problem because I need to use a Windows 3rd Lib, which uses WideString.

Then just use WideString or UnicodeString where needed. It is not a problem.

Note, WideString is for OLE programming. Most often you should use UnicodeString. Their memory management differs.

Juha

I cannot say from others, but I had this issue (about WideString) for now.

> This thread is not only about WinAPI. I have this problem because I > need to use a Windows 3rd Lib, which uses WideString. Then just use WideString or UnicodeString where needed. It is not a problem.

Are you saying that I need to do this?
(following the firt example on this thread)

=== begin ===
var
  U: UnicodeString;
  W: WideString;
begin
  U := IniFile.ReadString('TheLib', 'license', '');
  W := U;
  Lib.SetLicense(W);
  // ...
end;
=== end ===

...and I will not get a "Warning", right?

Note, WideString is for OLE programming. Most often you should use UnicodeString. Their memory management differs.

Ok... thanks... but in my case is a OLE object that I need to use.

Best regards,
Marcos Douglas

Marcos Douglas B. Santos wrote

I cannot say from others, but I had this issue (about WideString) for now.

The section "Calling Windows API" says:
'Only the "W" versions of Windows API functions should be called. It is like in Delphi except that you must assign strings to/from API calls to UnicodeString variables or typecast with UnicodeString().' Then it also explains the difference between WideString and UnicodeString. I should add a mention about PWideChar parameters. Anyway the idea is to keep the information useful and dense. Earlier it was bloated and intimidating.

Are you saying that I need to do this? (following the firt example on this thread)

No, if the parameter is WideString, not a pointer PWideChar, you can just call it like you did. Suppress the warning as Mattias told if it bothers you. You can also make a helper function so the conversion happens in one place.
Yes, for OLE you need WideString.

Juha

In a tight loop where one accesss the string character by character (take Pos() for example) this will lead to a significant slowdown as the compiler (without optimizations) will have to insert a call to the lookup function for each access. While I generally don't consider performance degradation as a backwards compatibility issue I do in this case, due to the significant decrease in performance.

Take this evaluation example:

=== code begin ===

program tperf;
{$mode objfpc}{$H+}
uses
  SysUtils;
function lookup(const aStr: String; aIndex: SizeInt): Char;
begin
  Result := aStr[aIndex];
end;
var
  str: String;
  starttime, endtime: TDateTime;
  i, j: LongInt;
begin
  SetLength(str, 10000);
  starttime := Now;
  for i := 0 to 10000 do
    for j := 1 to Length(str) do
      if str[j]  '' then ;
  endtime := Now;
  Writeln('Direct: ', FormatDateTime('hh:nn:ss.zzz', endtime - starttime));
  starttime := Now;
  for i := 0 to 10000 do
    for j := 1 to Length(str) do
      if lookup(str, j)  '' then ;
  endtime := Now;
  Writeln('Lookup: ', FormatDateTime('hh:nn:ss.zzz', endtime - starttime));
end.

=== code end ===

=== output begin ===

Direct: 00:00:01.766
Lookup: 00:00:02.061

=== output end ===

While this example is of course artificial it nevertheless shows the slow down.

Is your problem really more about the result type as, depending on the character width, the result could be an AnsiChar or WideChar or a UTF8 character for which I don't believe there is a defined char type (other than an arguable mis-use of UCS4Char)?

That is indeed also a problem. I might not have had that one in mind with my mail above, but I did back then when I had brainstormed this.

I can accept that a clear up of this area would also have to extend to the char types as well - but I would also argue that that is well overdue. On a quick count, I found 7 different char types in the system unit.

And most important of all: any solution that is developed *MUST* be backwards compatible, so that means that in the least that type aliases would remain anyway.

Regards,
Sven

You are wrong. The string types in 3.0.x and 3.1 are like this:

TP, Iso, ExtPas, MacPas, FPC, ObjFPC (or below modes with $H-): String = ShortString
Delphi (or other modes with $H+): String = AnsiString (or more precisely String(CP_ACP), meaning the system codepage)
Delphi_Unicode (or other modes with $H+ and $modeswitch unicodestring): String = UnicodeString

Regards,
Sven

El 16/08/17 a les 01:17, Graeme Geldenhuys via Lazarus ha escrit:

In hind sight, using TBytes or TMemoryStream and it would have been very clear that it is a storage container for byte sized data, and no automatic conversion (by the compiler) would be done to data stored in such containers.

Call me lazy but I don't want to reinvent the wheel and re-implement from scratch the functionality that a plain ansistring provides and TBytes to this day doesn't.
I mean, TBytes is just an "array of char". I can't (easily) add a byte to the end, cut a slice of the bytes, find one byte in the array, etc. OK, I can, but I have to program it all by myself while a string does all that and more and probably it's a lot more efficient.

Bye

El 16/08/17 a les 20:26, Luca Olivetti via Lazarus ha escrit:

Not to mention that its index starts from 0. If I wanted to program in C I would be programming in C, not pascal ;-)

Bye

Trunk supports Insert() and Delete() on dynamic arrays, Concat() and + are on the near term ToDo list.

Regards,
Sven

El 16/08/17 a les 22:40, Sven Barth via Lazarus ha escrit:

I started using strings as communication buffers since delphi 2. There weren't even dynamic arrays then...

Bye

Sven Barth wrote

You are wrong. The string types in 3.0.x and 3.1 are like this:

Thanks for correcting me. I was thinking of the "$modeswitch unicodestring" option.

Regards,
Graeme

Luca Olivetti wrote

I mean, TBytes is just an "array of char".

NO! Char can now mean a 1-byte char or a 2-byte char (I don't know how FPC plans to support Unicode surrogate pairs which will require 4-bytes). In the olden days (Delphi 7 and FPC 2.6.4) the Char type might always have meant 1-byte, but it doesn't necessarily these days.

TBytes has always been a container for Byte data.

Regards,
Graeme

Luca Olivetti wrote

I started using strings as communication buffers since delphi 2. There weren't even dynamic arrays then...

Well, Link-Lists existed from the beginning of time. I used them plenty in my TP days, and adding, inserting, indexing etc was pretty easy. Maybe programmers have just become spoilt over time with all the "out of the box" functionality and actually become lazy in coding.

Regards,
Graeme

Graeme Geldenhuys wrote

On 2017-08-16 18:35, Sven Barth via Lazarus wrote: > You are wrong. The string types in 3.0.x and 3.1 are like this: Thanks for correcting me. I was thinking of the "$modeswitch unicodestring" option.

will that modeswitch take care of the warning about explicit conversion between ansistring and unicode string when one has

var foo : unicodestring;
writeln(padright(foo,5);

??

i wrote a quick and simple little array exhibit program for someone... i had thought to try to embrace this new unicode stuff by using unicode strings... the using the padright and similar string manipulators gave me warnings about ansistring conversions :?

NOTE: this may be because i have an older lazarus and fpc installed... lazarus fixes 1.6.1 and fpc fixes 3.0.something...

Luca Olivetti wrote

I started using strings as communication buffers since delphi 2. There weren't even dynamic arrays then...

really? delphi came from TP/BP... i was (still am, actually) using dynamic arrays in TP6 ;)

Am 17.08.2017 04:16 schrieb "wkitty42--- via Lazarus" <

dynamic arrays in TP6 ;)

Dynamic arrays in the form of "array of Type" were only introduced in Delphi 3 if I remember correctly. Anything before that needed manual memory management.

Regards,
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170817/c359c76b/attachment.html>

Luca Olivetti wrote

Call me lazy but I don't want to reinvent the wheel and re-implement from scratch the functionality that a plain ansistring provides and TBytes to this day doesn't.

So please continue in the thread "dynamic string proposal".

Exactly this is part of what is discussed there.

-Michael

Sven Barth wrote

Trunk supports Insert() and Delete() on dynamic arrays, Concat() and + are on the near term ToDo list.

Supposedly "pos", as well. But that does not really help if we don't have a TStringList workalike, and supposedly several more library functions.

That is why I feel empowering the string paradigm for such use would be more appropriate. (See the thread "dynamic string proposal").

-Michael

Am 17.08.2017 11:21 schrieb "Michael Schnell via Lazarus" <

are on the near term ToDo list.

Supposedly "pos", as well. But that does not really help if we don't have

a TStringList workalike, and supposedly several more library functions.

That is why I feel empowering the string paradigm for such use would be

more appropriate. (See the thread "dynamic string proposal").

Why do you want to stuff everything and the kitchen sink into TStrings? There are much more suitable and less specialized container types available for this (to name a few: TFPGList, TList, etc.).

Regards,
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170817/e1f9a18f/attachment.html>

Luca Olivetti wrote

I started using strings as communication buffers since delphi 2. There weren't even dynamic arrays then... From the Turbo Pascal Help:

"A string type variable is a sequence of characters ..."

And then when you click on "characters":

"Char type
-----------
Variables of the ordinal type Char are used to store ASCII characters."

None of this suggests that string is a good type for storing arbitrary byte sequences.

You misused an implementation detail of the type (Ansi)String. And now you blame fpc.

You should have used a sane type for your buffer from the start.

Bart

I had D3 Pro, and this did definitively NOT support dynamic arrays. (Even String still was ShortString.)
All arrays had to be fixed range.
The often used construct to bypass this limitation was: Array[0..0] of TSomeType and have Range checking of.

Bart

You do like to use the word "ignorance" don't you. You can if you want take the view that all the "other programmers" that got the wrong idea are "stupid monkeys that don't know any better" or, alternatively, that they just wanted a nice cup of tea rather than the not quite tea drink that was served up.

Wikipedia sums the problem up nicely: "The early 2-byte encoding was usually called "Unicode", but is now called "UCS-2". UCS-2 differs from UTF-16 by being a constant length encoding and only capable of encoding characters of BMP, it is supported by many programs."

This is where the problem starts. The definitive of "Unicode" was changed (foolishly in my opinion) after it had been accepted by the community and the result is confusion. Hence my first point about not even using it. In using "UTF16/Unicode" I was attempting to convey the common use of the term which is to see UTF-16 as what is now defined as UCS-2. This is because hardly anyone I know uses UCS-2 and instead says "Unicode". Perhaps I just spend too much time amongst the ignorant.

Wikipedia also makes the wonderful point that "The UTF-16 encoding scheme was developed as a compromise to resolve this impasse in version 2.0". The impasse having resulted from "4 bytes per character wasted a lot of disk space and memory, and because some manufacturers were already heavily invested in 2-byte-per-character technology".

Finally: "In UTF-16, code points greater or equal to 2^16 are encoded using /two/ 16-bit code units. The standards organizations chose the largest block available of un-allocated 16-bit code points to use as these code units (since most existing UCS-2 data did not use these code points and would be valid UTF-16). Unlike UTF-8 they did not provide a means to encode these code points".

Which is from where I get my own view that UTF-16, as defined by the standards, is pointless. If you keep it to a UCS-2 (like) subset then you can get rapid indexing of character arrays. But as soon as you introduce the possibility of some characters being encoded as two 16-bit units then you lose rapid indexing and I can see no advantage over UTF-8
- plus you get all the fun of worrying about byte order.

Indeed, I believe those lazy programmers that you referred to, are actually making a conscious decision to prefer to work with a 16-bit code point only UTF-16 subset (i.e. the Basic Multilingual Plan) precisely so that they can do rapid indexing. As soon as you bring in 2 x 16-bit code unit code points, you lose that benefit - and perhaps you should be using UTF-32.

IMHO, Linux has got it right by using UTF-8 as the standard for character encoding and one of Lazarus's USPs is that it follows that lead - even for Windows. I can see why a program that does intensive text scanning will use a UTF-16 constrained to the BMP (i.e. 16-bit only), but not why anyone would prefer an unconstrained UTF-16 over UTF-8.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170817/7be1fb26/attachment.html>

Are you making my points for me? If such a basic term as "character" means 7 different things then something is badly amiss. It should be fairly obvious that in this context, character = printable symbol - whilst for practical reasons allowing for format control characters such as a "end of line" and "end of string".

I believe that you need to go back to the idea that you have both an abstract representation of a character with a constant semantic, separate from the actual encoding and for which there may be many different and valid encodings. For example, using a somewhat dated comparison, a lower case latin alphabet letter 'a' should always have a constant semantic, but in ASCII is encoded as decimal 97, while in EBCDIC is encoded as decimal 129. Even though they have different binary values, the represent the same abstract character.

I want a 'char' type in Pascal to represent a character such as a lower case 'a' regardless of the encoding used. Indeed, for a program to be properly portable, the programmer should not have to care are the actual encoding - only that it is a lower case 'a'.

Hence my proposal that a character type should include an implicit or explicit attribute that records the encoding scheme used - which could vary from ASCII to UTF-32.

You can then go on to define a text string as an array of characters with the same encoding scheme.

I was referring to GB 18030 and that it has one, two and four byte code points.

The point I believe that you are missing is to consider that a character is an abstract symbol with a semantic independent of how it is encoded. Collation sequences are independent of encoding and should remain the same regardless of how a character set is encoded.

Bart wrote

Variables of the ordinal type Char are used to store ASCII characters."

According to this wording, using Windows with ANSI character set would be a no-go.

-Michael

Tony Whyman wrote

UCS-2 differs from UTF-16 by being a constant length encoding and only capable of encoding characters of BMP, it is supported by many programs."

Rather obviously Embarcadero primarily had UCS-2 in mind as they created the "Unicode aware" Delphi. While it in fact does support full Unicode, keeping MyChar:=MyString[i] in place suggests to presume UCS-2 coded text for "unaware" programmers.

-Michael

Tony Whyman wrote

Finally: "In UTF-16, code points greater or equal to 2^16 are encoded using /two/ 16-bit code units.

2¹⁵ ???
-Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170817/adeebc33/attachment.html>

"Suppress the warning as Mattias told if it bothers you"

Of course bothers me.
Sorry, but every single warning is a... warning... that needs to be resolved. If this is not a problem (or a possible future problem), the compiler should not give us a warning, right?

Best regards,
Marcos Douglas

Am 17.08.2017 14:32 schrieb "Michael Schnell via Lazarus" <

a no-go.

Bart quoted from the TP help. And TP was written for DOS. There wasn't any Unicode or ANSI around yet...

Regards
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170817/09127ed9/attachment.html>

Am 17.08.2017 12:17 schrieb "Bart via Lazarus" <

memory

Then it was Delphi 4 ^^'

Regards,
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170817/3817dbd2/attachment-0001.html>

There was none at the start.

Martin

Marcos Douglas B. Santos wrote

Sorry, but every single warning is a... warning... that needs to be resolved.

I feel exactly the same. :-) It took me ages to figure out how to change my code so I could get rid of the "variable not initialized" whenever you used FillChar().

Regards,
Graeme

IIRC, we do it with

   var foo : array[0..0] of blerg;

and yes, we had/have to be very careful... i think range checking had to be off, also, because the additional elements were outside of the declared compile time range... we didn't have anything like setlength() to adjust the range at that time...

And what do you use? The Default() intrinsic function?

Ondrej

El 17/08/17 a les 01:34, Graeme Geldenhuys via Lazarus ha escrit:

On 2017-08-16 19:26, Luca Olivetti via Lazarus wrote: > I mean, TBytes is just an "array of char". NO! Char can now mean a 1-byte char or a 2-byte char (I don't know how

Sorry, I meant "array of byte". The point is it doesn't have all the features of a string.

Bye

Ondrej Pokorny wrote

> change my code so I could get rid of the "variable not initialized" > whenever you used FillChar(). And what do you use?

I implemented a FillMem() function that calls FillChar(). The parameter are differently defined. Then changed my code as follows:

  var
    msgp: TfpgMessageParams;
  begin
-  fillchar(msgp, sizeof(msgp), 0);
+  FillMem(@msgp, sizeof(msgp), 0);
    msgp.user.Param1 := 2;
    msgp.user.Param2 := x+10;

Finally that warning disappeared.

The FillMem() implementation is as follows:

procedure FillMem(Dest: pointer; Size: longint; Data: Byte);
begin
   FillChar( Dest^, Size, Data );
end;

Regards,
Graeme

Damn, should remember never to copy and paste from Wikipedia!

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170818/07067a7c/attachment.html>

Sven Barth wrote

Why do you want to stuff everything and the kitchen sink into TStrings?

To make use of the benefits the string type offers such as reference counting and lazy copy.

-Michael

Tony Whyman wrote

Damn, should remember never to copy and paste from Wikipedia!

Or simply use "plain text" emails to this mailing list. It reduces size considerably, and is still perfectly readable (like it has been for the last 30 years in email communications).

https://en.wikipedia.org/wiki/ASCII_Ribbon_Campaign

;-)

Regards,
Graeme

Graeme Geldenhuys wrote

On 2017-08-18 09:16, Tony Whyman via Lazarus wrote: > Damn, should remember never to copy and paste from Wikipedia! Or simply use "plain text" emails

To explain my mail above in plain Text:

A 32 bit Unicode needs two UTC-16 codes when it's greater than 2^15 (and not 2^16) , because bit 15 in UTF-16 is used to denote that another word follows.

-Michael

Michael Schnell wrote

To explain my mail above in plain Text:

Nice. :-)

And clever email clients will even "format" the plain text emails - looking much better than your original HTML version. See attached screenshot of how your last email looked like here.

Regards,
Graeme

Not exactly. ;-)
Please read the utf-16 specification again. The surrogate pair first code unit range is 0xDC00..0xDFFF. The rest can be used to encode code points of the BMP. There are no valid code points in range 0xDC00..0xDFFF. https://en.wikipedia.org/wiki/UTF-16

Martin

Am 18.08.2017 10:42 schrieb "Michael Schnell via Lazarus" <

counting and lazy copy.

The same is true for dynamic arrays.
So again the question where is the need for TStrings? And don't come along with the String operations like Delete(), Insert() and Pos() again, because those aren't covered by TStrings either.

Regards,
Sven
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus-ide.org/pipermail/lazarus/attachments/20170818/7c5bbb71/attachment.html>

Sven Barth wrote

Delete(), Insert() and Pos()

I understand those, and "+", etc, are in the pipe for Array of Byte, as well.

I suppose workalikes for the most important TStrings siblings, such as TStringlist (including sort and LoadFromFile) be doable on top of that.

Might be good news for new programs. Legacy stuff will need to be ported appropriately (as already discussed here).

-Michael

Sven Barth wrote

Trunk supports Insert() and Delete() on dynamic arrays, Concat() and + are on the near term ToDo list.

I eagerly wait for these (and for anonymous methods) :)

Ondrej