Start to fusl toolchain

Start to fusl toolchain

R=viettrungluu@chromium.org

Committed: https://chromium.googlesource.com/external/mojo/+/006e65581006bc0328bc0c5c184587b4cd41f75c

[kulakowski, 2016-01-29 23:46:32]

Description was changed from

==========
Start to fusl toolchain

BUG=
==========

to

==========
Start to fusl toolchain
==========

[kulakowski, 2016-01-29 23:46:32]

kulakowski@chromium.org changed reviewers:
+ phosek@chromium.org, viettrungluu@chromium.org

[kulakowski, 2016-01-29 23:49:16]

On 2016/01/29 23:46:32, kulakowski wrote:
> mailto:kulakowski@chromium.org changed reviewers:
> + mailto:phosek@chromium.org, mailto:viettrungluu@chromium.org

This works. The hack around crt linker stuff mentioned in BUILD.gn is annoying,
but the dummy action did the trick (thanks Trung).

However, this currently depends on populating the last bits of the C runtime,
the bits that come from the compiler rather than the libc. If I manually copy
over crtbegin.o, crtend.o, and libgcc.a, and make an empty libgcc_s.a, then
empty_main builds and runs.

Petr, you'd mentioned knowing how to get clang to dump the location of those
files. I can figure it out if you have other things ongoing.

[Petr Hosek, 2016-01-30 00:02:56]

https://codereview.chromium.org/1649133002/diff/40001/fusl/BUILD.gn
File fusl/BUILD.gn (right):

https://codereview.chromium.org/1649133002/diff/40001/fusl/BUILD.gn#newcode1588
fusl/BUILD.gn:1588: # etc. from ending up in downstream targets' linker response
files.
That's what data_deps is for, no need for Python script or custom action.

[kulakowski, 2016-01-30 00:04:45]

https://codereview.chromium.org/1649133002/diff/40001/fusl/BUILD.gn
File fusl/BUILD.gn (right):

https://codereview.chromium.org/1649133002/diff/40001/fusl/BUILD.gn#newcode1588
fusl/BUILD.gn:1588: # etc. from ending up in downstream targets' linker response
files.
On 2016/01/30 00:02:56, Petr Hosek wrote:
> That's what data_deps is for, no need for Python script or custom action.

There was some discussion in the mojo irc about this. It really is a build time
dependency.

The problem is that there is a transitive dependency from empty_main (my test
program) back to crt1.c, and so gn injects crt1.o into the linker script. This
then blows up with duplicate _init and _start and whatnot.

[kulakowski, 2016-01-30 01:14:00]

On 2016/01/30 00:04:45, kulakowski wrote:
> https://codereview.chromium.org/1649133002/diff/40001/fusl/BUILD.gn
> File fusl/BUILD.gn (right):
> 
>
https://codereview.chromium.org/1649133002/diff/40001/fusl/BUILD.gn#newcode1588
> fusl/BUILD.gn:1588: # etc. from ending up in downstream targets' linker
response
> files.
> On 2016/01/30 00:02:56, Petr Hosek wrote:
> > That's what data_deps is for, no need for Python script or custom action.
> 
> There was some discussion in the mojo irc about this. It really is a build
time
> dependency.
> 
> The problem is that there is a transitive dependency from empty_main (my test
> program) back to crt1.c, and so gn injects crt1.o into the linker script. This
> then blows up with duplicate _init and _start and whatnot.

PTAL

1. The sysroot is now done, so the build should actually work and the bot be
green.

2. Adding the action to copy the host system CRT stuff supplants the need for
the dummy action, so this got cleaner.

[viettrungluu, 2016-02-01 21:15:55]

nits

https://codereview.chromium.org/1649133002/diff/100001/fusl/BUILD.gn
File fusl/BUILD.gn (right):

https://codereview.chromium.org/1649133002/diff/100001/fusl/BUILD.gn#newcode1473
fusl/BUILD.gn:1473: static_library("libgcc_s") {
What's this for?

https://codereview.chromium.org/1649133002/diff/100001/fusl/BUILD.gn#newcode1477
fusl/BUILD.gn:1477: static_library("libm") {
I assume this is so "-lm" works? Could you add a comment to that effect?

https://codereview.chromium.org/1649133002/diff/100001/fusl/test/empty_main.c
File fusl/test/empty_main.c (right):

https://codereview.chromium.org/1649133002/diff/100001/fusl/test/empty_main.c...
fusl/test/empty_main.c:5: int main(void) {
Dunno if "void" is actually correct/more acceptable (should be "int,
char**")....

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
File fusl/tools/populate_crt.py (right):

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
fusl/tools/populate_crt.py:18: 'crtbegin.o',
I think we mostly format Python code with a 2-space indent (which is a crock,
but what can you do).

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
fusl/tools/populate_crt.py:56: # Check that everything is here.
Is it possible to find some things, but not all? (Wouldn't the search rules be
that it grabs each thing from the first thing in the search path that has it?)

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
fusl/tools/populate_crt.py:57: for artefact in SYSROOT_ARTEFACTS:
In 'merika, "we" spell artifact with an "i".

[kulakowski, 2016-02-01 21:29:29]

https://codereview.chromium.org/1649133002/diff/100001/fusl/BUILD.gn
File fusl/BUILD.gn (right):

https://codereview.chromium.org/1649133002/diff/100001/fusl/BUILD.gn#newcode1473
fusl/BUILD.gn:1473: static_library("libgcc_s") {
On 2016/02/01 21:15:55, viettrungluu wrote:
> What's this for?

Will add comment.

https://codereview.chromium.org/1649133002/diff/100001/fusl/BUILD.gn#newcode1477
fusl/BUILD.gn:1477: static_library("libm") {
On 2016/02/01 21:15:55, viettrungluu wrote:
> I assume this is so "-lm" works? Could you add a comment to that effect?

Ditto.

https://codereview.chromium.org/1649133002/diff/100001/fusl/test/empty_main.c
File fusl/test/empty_main.c (right):

https://codereview.chromium.org/1649133002/diff/100001/fusl/test/empty_main.c...
fusl/test/empty_main.c:5: int main(void) {
On 2016/02/01 21:15:55, viettrungluu wrote:
> Dunno if "void" is actually correct/more acceptable (should be "int,
> char**")....

I clearly don't care for this, but for the sake of pedantry, (void) is allowed,
as is (int, char**) or equivalent, as is anything the implementation chooses to
define.

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
File fusl/tools/populate_crt.py (right):

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
fusl/tools/populate_crt.py:56: # Check that everything is here.
On 2016/02/01 21:15:55, viettrungluu wrote:
> Is it possible to find some things, but not all? (Wouldn't the search rules be
> that it grabs each thing from the first thing in the search path that has it?)

Ah, yeah. Didn't think too much about it, since in practice it won't happen.
Will fix.

https://codereview.chromium.org/1649133002/diff/100001/fusl/tools/populate_cr...
fusl/tools/populate_crt.py:57: for artefact in SYSROOT_ARTEFACTS:
On 2016/02/01 21:15:55, viettrungluu wrote:
> In 'merika, "we" spell artifact with an "i".

Done.

[kulakowski, 2016-02-01 22:41:30]

I changed all the stuff, and added a big comment about what the different bits
of the C runtime are.

I also fixed the linker invocation to be static, so it links in the correct
stuff.

I also copied the rest of the C runtime, including bits we currently aren't
using (for dynamic linking etc.).

[viettrungluu, 2016-02-01 23:52:57]

lgtm

[kulakowski, 2016-02-02 02:10:31]

Description was changed from

==========
Start to fusl toolchain
==========

to

==========
Start to fusl toolchain

R=viettrungluu@chromium.org

Committed:
https://chromium.googlesource.com/external/mojo/+/006e65581006bc0328bc0c5c184...
==========

[kulakowski, 2016-02-02 02:10:32]

Committed patchset #8 (id:140001) manually as
006e65581006bc0328bc0c5c184587b4cd41f75c (presubmit successful).