[fpc-devel] Building trunk on Win32

[Sven Barth]

On 27.11.2011 18:48, Sergei Gorelkin wrote:
> 27.11.2011 19:10, Sven Barth пишет:
>> On 27.11.2011 16:45, Sergei Gorelkin wrote:
>>> The fact it works with DISABLE_TLS_DIRECTORY and fails otherwise
>>> suggests that the foreign thread raises exceptions and FPC tries to
>>> handle them. Error 255 is most likely caused by Halt(255) in
>>> rtl/inc/except.inc line 200 or 303, these are the only places where RTL
>>> can exit with code 255.
>>> This is explainable: FPC won't have ExceptObjectStack or
>>> ExceptAddressStack set up for the foreign thread, they will contain nil
>>> values.
>>>
>>> But I don't imagine how it manages to work *without* TLS callbacks. The
>>> difference is that RTL will remain in single-threaded mode, but
>>> SetUnhandledExceptionFilter that is used to intercept exceptions is
>>> AFAIK global and will still break in.
>>
>> Also interesting question: why does so far only Leonardo experience
>> that problem? I myself have no
>> problem building trunk on a Windows 7 machine using 2.4.4 as a
>> starting compiler.
>>
> He has some software which creates threads in compiler process. Maybe
> something specific to VM, maybe something specific to server OS.
> I also build trunk successfully on Windows XP (32bit) and Windows 7
> (64bit), on both of them Process Explorer shows just a single thread in
> compiler process.
>
> Another point to look at is webtbs/tw2423.pp, testsuite shows it fails
> consistently with RTE 226, while I have never seen it failing locally.

I've remembered that I also have a Windows 2003 VM running on my Linux 
x86_64 system (using QEMU though). I did now run a "make cycle" and 
tested the test you mentioned (using the newly compiled compiler). No 
problems here (revision is 19692 - it's a coincidence though that it's 
directly before your change). I also checked the compiler process using 
Process Explorer: only one thread there.

Regards,
Sven