[cfe-dev] Merging MinGW toolchain patches to 3.7

[Yaron Keren]

Hi Richard,

Both issues implemented in r243098, r243101, r243105 per your suggestions.

It may be of general interest that two builders: sanitizer-x86_64-linux-fast
 and clang-x86_64-ubuntu-gdb-75 add .svn entries in some (all?) directories
which may cause code that does not expect them there to malfunction.

Yaron


2015-07-24 2:25 GMT+03:00 Richard Smith <richard at metafoo.co.uk>:

> On Thu, Jul 23, 2015 at 2:56 PM, Yaron Keren <yaron.keren at gmail.com>
> wrote:
>
>> Hi Richard,
>>
>> r242660 + r242666 + r242667 are the current mingw toolchain, should be
>> tested together.
>> The code test covarage is not complete:
>>
>> 1) To make the toolchain work automatically, the first stage is detection
>> of the mingw installation dir in MinGW::findGccLibDir(). This part is not
>> covered by any tests as it probes several file system directories outside
>> the LLVM tree using llvm::sys::fs::directory_iterator, we can't control
>> that using test infrastructure.
>>
>
> Please take a look at how other targets handle this (see the various
> sysroot trees under test/Driver/Inputs). If the same approach can't work
> here, we should figure out how to fix that. The MinGW driver support should
> be testable without having a MinGW toolchain installed, otherwise the cost
> imposed on people maintaining the driver and Clang in general is too high.
>
> The most reliable way to test this toolchain would be to have a mingw
>> toolchain installed on various bots and run tests using it. I'm open to
>> suggestions how to test MinGW::findGccLibDir() otherwise.
>>
>> 2) The second stage generates the C and C++ include directories based
>> on GccLibDir found in stage one. This *is* covered by the tests introduced
>> in r242666 by injecting GccLibDir with --sysroot argument (instead of
>> probing the file system) for both Windows and Linux.
>>
>> MinGW::Arch has a trainling slash as a small optimization so Arch could
>> be concatenated into a path instead of using sys::path::append. If you find
>> that confusing I can remove the trailing slash and use sys::path::append
>> calls instead, it would not matter much.
>>
>
> There should at least be a comment explaining this in the class
> definition. But generally, using the higher-level facilities such as
> sys::path::append should be preferred over manually inserting separators
> into paths, so I'd prefer you did that. Performance microoptimizations are
> not especially relevant for the driver, since it's such a tiny proportion
> of the cost of a compilation.
>
>
>> Yaron
>>
>>
>>
>>
>> 2015-07-24 0:09 GMT+03:00 Richard Smith <richard at metafoo.co.uk>:
>>
>>> On Thu, Jul 23, 2015 at 12:30 PM, Richard Smith <metafoo at gmail.com>
>>> wrote:
>>>
>>>> Yes, we should get these changes into 3.7. I'll look over the patches
>>>> later today.
>>>>
>>> I'm not very happy about applying patches with no tests to the branch,
>>> but, reluctantly, approved.
>>>
>>> Yaron: please add some tests for r242660 and r242667 to trunk. (I also
>>> find it very weird that the MinGW::Arch member has a trailing slash. Is
>>> there a good reason for that?)
>>>
>>>> On 23 Jul 2015 9:56 am, "Hans Wennborg" <hans at chromium.org> wrote:
>>>>
>>>>> +cfe-dev to keep it on the list, and Richard
>>>>>
>>>>> This sounds like a reasonable thing to merge.
>>>>>
>>>>> Richard, what's your opinion as an owner?
>>>>>
>>>>>
>>>>> On Thu, Jul 23, 2015 at 9:48 AM, Martell Malone <
>>>>> martellmalone at gmail.com> wrote:
>>>>> > I almost forgot
>>>>> >
>>>>> > Yaron also fixed all the different combination of linux host/distros
>>>>> > targetting mingw with theses 3.
>>>>> > I have a feeling we will have people hitting the mailing list about
>>>>> come the
>>>>> > 3.7 release if their distro isn't working.
>>>>> >
>>>>> > clang
>>>>> >
>>>>> > r242660  Support mingw toolchain include and lib directories on Arch
>>>>> Linux.
>>>>> > r242667  Remove erroneous space in "lib64" string constant.
>>>>> > r242766  Fix mingw toolchain to honor sysroot on Linux and add tests.
>>>>> >
>>>>> > Again as i said I've no idea what the process is or what we can do
>>>>> here.
>>>>> > Hoping you can fill me in on the details
>>>>> >
>>>>> > Kind Regards
>>>>> > Martell
>>>>> >
>>>>> >
>>>>> >
>>>>> > On Thu, Jul 23, 2015 at 5:44 PM, Martell Malone <
>>>>> martellmalone at gmail.com>
>>>>>
>>>>> > wrote:
>>>>> >>
>>>>> >> Hi Hans :)
>>>>> >>
>>>>> >> rnk and I were talking to me about the possibility of getting the
>>>>> mingw
>>>>> >> driver work back ported into the 3.7 branch.
>>>>> >>
>>>>> >> I read the thread where you branched the 3.7 release you said that
>>>>> if
>>>>> >> anyone has a request to have patches merged they should contact you
>>>>> directly
>>>>> >>
>>>>> >> When the branch was made we were in the middle of crushing the
>>>>> final bugs
>>>>> >> in the MINGW Driver.
>>>>> >> I had a total of about 9 patches merged but 2 missed the window
>>>>> >> unfortunately.
>>>>> >> 1 was for compiler-rt and 1 was for clang
>>>>> >>
>>>>> >> clang
>>>>> >> r242905  Add support for -rtlib option and -stdlib option to the
>>>>> mingw
>>>>> >> driver
>>>>> >>
>>>>> >> compiler-rt
>>>>> >> r242539  compiler-rt: add support for mingw-w64 in builtins
>>>>> >> r242540  Add missing chkstk.S files from r242539
>>>>> >>
>>>>> >> (r242540 is here because only half the patch was committed
>>>>> initially)
>>>>> >>
>>>>> >> These 2(now 3) patches complete my series and fixes everything to
>>>>> have a
>>>>> >> clang toolchain with mingw-w64 without needing gcc.
>>>>> >>
>>>>> >> I don't know what the process is for back porting or how you feel
>>>>> about
>>>>> >> these patches but I said I would email you about it an see where we
>>>>> can go
>>>>> >> from here :)
>>>>> >>
>>>>> >> Kind Regards
>>>>> >> Martell
>>>>> >
>>>>> >
>>>>>
>>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150724/92b5cc05/attachment.html>