[llvm] r200340 - Add line table debug info to COFF files when using a win32 triple.

[Timur Iskhodzhanov]

This got reverted as r200375.

28 янв. 2014 г. 22:23 пользователь "NAKAMURA Takumi" <geek4civic at gmail.com>
написал:

Reverted in r200375. It broke many llvm and clang tests, at least in msc17
> builder.
>

It would help a bit if you also gave a link to the report.


> ================
> Comment at: lib/CodeGen/AsmPrinter/AsmPrinter.cpp:198
> @@ -195,3 +197,3 @@
>    if (MAI->doesSupportDebugInformation()) {
> -    DD = new DwarfDebug(this, &M);
> -    Handlers.push_back(HandlerInfo(DD, DbgTimerName, DWARFGroupName));
> +    if (Triple(TM.getTargetTriple()).getOS() == Triple::Win32) {
> +      Handlers.push_back(HandlerInfo(new WinCodeViewLineTables(this),
> ----------------
> You should tweak more other tests (also in clang/test/CodeGen/ !) when you
> suppressed llvm::DwarfDebug.
>
> You can reproduce such a case with
> LLVM_DEFAULT_TARGET_TRIPLE:=i686-pc-win32
>
>
> http://llvm-reviews.chandlerc.com/D2232




> On Tue, Jan 28, 2014 at 10:54 PM, Timur Iskhodzhanov <timurrrr at google.com>
>  wrote:
>
>> Sorry for the breakage and not reverting it earlier - I don't think I got
>> an email about the failure.
>>
>> How do you recommend me to tweak them?
>> The default triple on windows shouldn't imply dwarf. I think running
>> these tests with a -cygwin/-mingw32 triple is reasonable, as it is probably
>> (?) reasonable to run them with the default triple on non-Windows
>> architectures.
>>
> 2014-01-29 Reid Kleckner <rnk at google.com>

> I'm pretty sure Timur is testing with a triple of i686-pc-win32.
>

It's a bit more complex than that: I only ran new tests on i686-pc-win32 as
I don't have the workstation fully set up.
I ran the whole suite on my Mac laptop expecting it to have enough
coverage. Oh, well.


> One way to tweak the tests is to use -triple %itanium_abi_triple, although
> that's abusing the terminology a little bit.  It's better than using an
> explicit triple because you don't over specify the test requirements.
>

I've looked at the failures and most of them seem to have an implicit
triple.

I'm pretty sure we want to exclude some tests entirely for Win runs.
For example, it's almost certainly wrong to run MIPS or SPARC tests on
Windows with an implicit triple?

For the other tests, I plan to add something like "-triple %dwarf_triple"
that works similar to %itanium_abi_triple, i.e. replace win32 with mingw32.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140129/27c0e988/attachment.html>