Don't use depfiles for assembly with Clang

Don't use depfiles for assembly with Clang

Clang doesn't seem to support depfiles for assembly leading to stream
of warnings when compiling fusl.

BUG=#619
R=kulakowski@chromium.org, viettrungluu@chromium.org

Committed: https://chromium.googlesource.com/external/mojo/+/0d2a6f788aafe8f53e24149d955d75fb1940d074

[viettrungluu, 2016-01-13 00:31:54]

Doesn't this result in not getting proper dependencies when building with clang?
Should we be using another tool to generate the depfiles?

[Petr Hosek, 2016-01-13 00:41:23]

On 2016/01/13 00:31:54, viettrungluu wrote:
> Doesn't this result in not getting proper dependencies when building with
clang?
> Should we be using another tool to generate the depfiles?

You're not suppose to be using C preprocessor on Assembly, C standard doesn't
say what's going to happen if you do. That being said, many existing libraries
like glibc or newlib rely on that and GCC will happily carry on which could (and
does) lead to errors when you try to compile same codebases with Clang which
doesn't officially support this use case (and likely never will).

I've checked musl assembly files and they don't rely on C preprocessor so that
answer is (fortunately for us) no and no.

[kulakowski, 2016-01-13 00:44:30]

On 2016/01/13 00:41:23, Petr Hosek wrote:
> On 2016/01/13 00:31:54, viettrungluu wrote:
> > Doesn't this result in not getting proper dependencies when building with
> clang?
> > Should we be using another tool to generate the depfiles?
> 
> You're not suppose to be using C preprocessor on Assembly, C standard doesn't
> say what's going to happen if you do. That being said, many existing libraries
> like glibc or newlib rely on that and GCC will happily carry on which could
(and
> does) lead to errors when you try to compile same codebases with Clang which
> doesn't officially support this use case (and likely never will).
> 
> I've checked musl assembly files and they don't rely on C preprocessor so that
> answer is (fortunately for us) no and no.

sgtm, also just checked and saw no #include #define #if etc.

lgtm

[Petr Hosek, 2016-01-13 01:07:57]

On 2016/01/13 00:44:30, kulakowski wrote:
> On 2016/01/13 00:41:23, Petr Hosek wrote:
> > On 2016/01/13 00:31:54, viettrungluu wrote:
> > > Doesn't this result in not getting proper dependencies when building with
> > clang?
> > > Should we be using another tool to generate the depfiles?
> > 
> > You're not suppose to be using C preprocessor on Assembly, C standard
doesn't
> > say what's going to happen if you do. That being said, many existing
libraries
> > like glibc or newlib rely on that and GCC will happily carry on which could
> (and
> > does) lead to errors when you try to compile same codebases with Clang which
> > doesn't officially support this use case (and likely never will).
> > 
> > I've checked musl assembly files and they don't rely on C preprocessor so
that
> > answer is (fortunately for us) no and no.
> 
> sgtm, also just checked and saw no #include #define #if etc.
> 
> lgtm

I guess one question is whether we should make it conditional for clang only as
it is now or make the toolchain setup the same for both gcc and clang?

[kulakowski, 2016-01-13 01:09:21]

On 2016/01/13 01:07:57, Petr Hosek wrote:
> On 2016/01/13 00:44:30, kulakowski wrote:
> > On 2016/01/13 00:41:23, Petr Hosek wrote:
> > > On 2016/01/13 00:31:54, viettrungluu wrote:
> > > > Doesn't this result in not getting proper dependencies when building
with
> > > clang?
> > > > Should we be using another tool to generate the depfiles?
> > > 
> > > You're not suppose to be using C preprocessor on Assembly, C standard
> doesn't
> > > say what's going to happen if you do. That being said, many existing
> libraries
> > > like glibc or newlib rely on that and GCC will happily carry on which
could
> > (and
> > > does) lead to errors when you try to compile same codebases with Clang
which
> > > doesn't officially support this use case (and likely never will).
> > > 
> > > I've checked musl assembly files and they don't rely on C preprocessor so
> that
> > > answer is (fortunately for us) no and no.
> > 
> > sgtm, also just checked and saw no #include #define #if etc.
> > 
> > lgtm
> 
> I guess one question is whether we should make it conditional for clang only
as
> it is now or make the toolchain setup the same for both gcc and clang?

Are there many other clients of this code? I don't recall many .s files
elsewhere.

[Petr Hosek, 2016-01-13 01:27:42]

On 2016/01/13 01:09:21, kulakowski wrote:
> On 2016/01/13 01:07:57, Petr Hosek wrote:
> > On 2016/01/13 00:44:30, kulakowski wrote:
> > > On 2016/01/13 00:41:23, Petr Hosek wrote:
> > > > On 2016/01/13 00:31:54, viettrungluu wrote:
> > > > > Doesn't this result in not getting proper dependencies when building
> with
> > > > clang?
> > > > > Should we be using another tool to generate the depfiles?
> > > > 
> > > > You're not suppose to be using C preprocessor on Assembly, C standard
> > doesn't
> > > > say what's going to happen if you do. That being said, many existing
> > libraries
> > > > like glibc or newlib rely on that and GCC will happily carry on which
> could
> > > (and
> > > > does) lead to errors when you try to compile same codebases with Clang
> which
> > > > doesn't officially support this use case (and likely never will).
> > > > 
> > > > I've checked musl assembly files and they don't rely on C preprocessor
so
> > that
> > > > answer is (fortunately for us) no and no.
> > > 
> > > sgtm, also just checked and saw no #include #define #if etc.
> > > 
> > > lgtm
> > 
> > I guess one question is whether we should make it conditional for clang only
> as
> > it is now or make the toolchain setup the same for both gcc and clang?
> 
> Are there many other clients of this code? I don't recall many .s files
> elsewhere.

Nope, it seems like the only Assembly code in Mojo is go, Android, Native Client
or fusl, so we may as well make unconditional.

[Petr Hosek, 2016-01-13 02:20:14]

Done, I've made the change non-conditional.

[Petr Hosek, 2016-01-13 17:55:42]

Description was changed from

==========
Don't use depfiles for assembly with Clang

Clang doesn't seem to support depfiles for assembly leading to stream
of warnings when compiling fusl.

BUG=#619
R=kulakowski@chromium.org, viettrungluu@chromium.org
==========

to

==========
Don't use depfiles for assembly with Clang

Clang doesn't seem to support depfiles for assembly leading to stream
of warnings when compiling fusl.

BUG=#619
R=kulakowski@chromium.org, viettrungluu@chromium.org

Committed:
https://chromium.googlesource.com/external/mojo/+/0d2a6f788aafe8f53e24149d955...
==========

[Petr Hosek, 2016-01-13 17:55:43]

Committed patchset #2 (id:20001) manually as
0d2a6f788aafe8f53e24149d955d75fb1940d074 (presubmit successful).