NaCL doesn't forget goma

[Sunny Sachanandani]

Hi,

I've been having issues with my computer hanging during a chrome build with goma on Linux. So I decided to try a local build but for some reason several NaCL related build actions complain about the goma socket being missing.

I've tried deleting the out/ directory and recreating the gyp files but that doesn't work either. Anyone know what's up?

The steps that fail look like this:

[119/19226] ACTION nacl_lib_newlib: build newlib x86-64 nlib_34595b1baf6b0067c0d7da5211c629a9

GOMA: port request failed. connect_success: 0, err: -1 Failed to connect to /tmp/goma.ipc, http_return_code: 0

Thanks!

- Sunny

[Thiago Farina]

On Fri, Jul 31, 2015 at 9:54 PM, Sunny Sachanandani <sun...@chromium.org> wrote:

Hi,

I've been having issues with my computer hanging during a chrome build with goma on Linux. So I decided to try a local build but for some reason several NaCL related build actions complain about the goma socket being missing.

Do you have goma_dir and use_goma set?

--

Thiago Farina

[Yoshisato Yanagisawa]

tl;dr please set NO_NACL_GOMA=true.  Sorry for annoying messages.

The hacks in build_nexe.py GetGomaConfig may do something bad for you.  If either of followings are set, build_nexe.py try to use gomacc.  Goma is used only when it confirms compiler_proxy available but you may see annoying messages shown by the checks.

1. use_goma is set in GYP_DEFINES environment variable or something equivalent.

2. GOMA_DIR environment variable is set.

3. gomacc can be found in PATH environment variable.

As you can see very beginning of the function, NO_NACL_GOMA=true kills the feature.

--
--
Chromium Developers mailing list: chromi...@chromium.org
View archives, change email options, or unsubscribe:
http://groups.google.com/a/chromium.org/group/chromium-dev

[Sunny Sachanandani]

I have the GOMA_DIR environment variable set. I'm assuming build_nexe.py works this way because it's used outside of building chromium so it can't just rely on the use_goma gyp variable?

[Yoshisato Yanagisawa]

Just curious but why you set GOMA_DIR when you do not use goma?

Anyway, will you file an issue for this?  GOMA_DIR support is provided for builds that set GOMA_DIR when they use goma (maybe clank and chromeos?).  We should confirm they do not depend on GOMA_DIR.  Also, we should confirm nobody use "masquerade mode" gomacc with build_nexe.py.

[Sunny Sachanandani]

I do use goma most of the time but I was debugging an issue where goma would cause my computer to hangup. I use the "use_goma" gyp variable to turn goma on or off as needed. I wasn't aware that just setting GOMA_DIR could cause goma to be used in build_nexe.py.

Filed a bug here: crbug.com/516552

Thanks!

[Paweł Hajdan, Jr.]

Just wondering, why don't we add a command-line flag to build_nexe.py to control whether it uses goma or not, and make GYP/GN pass that command-line flag or not based on use_goma flag?

I think that'd be more consistent, predictable and easier to reason about than some tools having surprising "heuristics". There could have been legitimate reasons to add that in the past - we could consider revisiting that though.

Paweł

[Yoshisato Yanagisawa]

2015-08-04 17:56 GMT+09:00 Paweł Hajdan, Jr. <phajd...@chromium.org>:

Just wondering, why don't we add a command-line flag to build_nexe.py to control whether it uses goma or not, and make GYP/GN pass that command-line flag or not based on use_goma flag?

Sorry for my bad explanation but it does.  If use_goma gyp varialbe is set, gyp/gn set --gomadir flag.  I hope we can be free from build_nexe.py with GN.

However, build_nexe.py also accepts GOMA_DIR and PATH in addition to --gomadir flag.  GOMA_DIR support is for clank and PATH is for masquerade mode.  I suppose both cases are not used anymore.

If my understanding is wrong, trybots should say so.  Let's see what happens with this:

https://codereview.chromium.org/1256313004

[Yoshisato Yanagisawa]

FYI, submitted https://codereview.chromium.org/1256313004.

GOMA_DIR and gomacc in PATH won't be used as a signal for use of goma when the change is rolled.

[Dongseong Hwang]

Since 11th/Aug, gomacc makes external contributors unhappy.

Even if I define use_goma=0, nacl tries to use gomacc, so build fails.

/home/dshwang/goma/gomacc ../native_client/toolchain/linux_x86/pnacl_newlib/bin/pnacl-clang -c ../out_ozone_cros/Release/gen/mojo/nacl/mojo_irt.c ....

Can someone fix it?

- DS

[Primiano Tucci]

Can you please file a bug on crbug and add Cr-Platform-NaCl and Build labels?

[Dongseong Hwang]

I filed the issue in https://code.google.com/p/chromium/issues/detail?id=520051