So I was doing some refactoring work, and ran the test suites to see if nothing broke.  It turned out that I got a single failure in test/cg/variants/tnofalvarol.pp - seems a little odd because my changes weren't meant to change the output of binary files in any way.  But looking at the test in question, all I get is a wall of $include definitions.  Running it in Lazarus isn't too helpful either because the stack trace doesn't reveal where the erroneous line is, only that seven levels down, an error occurs in "destructor tdynarrayiter.done;" in the variants unit, but doesn't know which line exactly - exception EVariantError with message: "Invalid variant type cast".

I suppose what I'm trying to get at is that this test is not very self-contained and is very hard to debug if it happens to fail, and I would consider rewriting or replacing it.

Gareth aka. Kit

---
This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus

"destructor tdynarrayiter.done;" is part of
packages/rtl-objpas/src/inc/variants.pp. You don't get more information because all units from the packages directory are compiled with optimizations and without debug information by default. You will have to compile that unit with debug information (you can add OPT="-O- -gw" to the make command when building all of FPC to get everything built without optimizations and with debug information).

I suppose what I'm trying to get at is that this test is not very self-contained and is very hard to debug if it happens to fail, and I would consider rewriting or replacing it.

The files it includes have been automatically generated, automatically tested against Kylix, and then
a) if the program compiled with Kylix, it was run to see which overload Kylix picked and whether it generated an exception, and a "halt(1)" was (also automatically) placed in the other paths. All successfully executing tests are included in tnofalvarol.pp to make the test go faster (compiling and running one program with 100 include files is much faster than compiling 100 separate files)
b) if the program did not compile with Kylix, it was kept as a separate program (since if you include two failing programs in a single program and it fails to compile, we can't test whether both errors were detected).

You can still compile the individual include files of the test as separate programs to test and debug them separately. Once you compile the variants unit with debug information, you should be able to easily see which individual test is crashing based on the backtrace.

Jonas

Thanks Jonas.

---
This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus

So I did try building it manually with the trunk, and for some reason it failed on that as well, even though it didn't appear in the test suite as a failure.  Definitely confusing, but it tells me that the failure might not be due to my changes.  I'll keep an eye out though to see if anything crops up.

Gareth aka. Kit