RFC: "analyze" for compile-only builders with targets like chromium_builder_tests

[Paweł Hajdan, Jr.]

The previous thread about "analyze" resulted in a nice plan to simplify the code and only use targets/matching_exes (https://groups.google.com/a/chromium.org/d/msg/infra-dev/3kUd4emLyBA/zPCnKUgk39MJ).

I'm wondering though what to do about compile-only builders, say "chromeos_x86-generic_chromium_compile_only_ng". Currently they compile "all". Even if we switch them to "chromium_builder_tests", it seems "analyze" would always pick that target anyway.

Maybe it's not so bad - ninja will only rebuild files that need rebuilding.

There is some chance we could avoid rebuilding files that are technically stale, but not affected by the patch.

Should we have different modes for "analyze"? How would this work with GN which for now seems to only have the "first" mode described at the beginning of this email?

Paweł

[Scott Violet]

This is why analyze has the two outputs. For build only bots if you
compile build_targets that should have the same effect as building
all.

-Scott

> --
> You received this message because you are subscribed to the Google Groups
> "infra-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to infra-dev+...@chromium.org.
> To post to this group, send email to infr...@chromium.org.
> To view this discussion on the web visit
> https://groups.google.com/a/chromium.org/d/msgid/infra-dev/CAATLsPb51DRrF7aNS7253LBa2JSuGyqhdRAG_nty8M7ehJOPAg%40mail.gmail.com.

[Dirk Pranke]

If we needed to, we could add the other mode to GN as well.

Or, we could have a "hybrid" model where if the input list of targets

contained groups, we would only include the targets in the group that

are actually affected.

I don't know that I have a strong leaning here myself. 

On Wed, Jul 1, 2015 at 1:48 PM, Scott Violet <s...@chromium.org> wrote:

This is why analyze has the two outputs. For build only bots if you
compile build_targets that should have the same effect as building
all.

You mean "same effect" in the sense that you'd rebuild all of the targets

affected by the patch, right? I would expect building "all" to also cause you

to rebuild targets that were out-of-date but otherwise unaffected by the patch

(they are just out of date because the build was otherwise out-of-date).

-- Dirk

 

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/infra-dev/CAKARY_%3D3gs4rE1u_NnDu8J%2Br9OiboJie_2gr3-aGaBr-uyvgDg%40mail.gmail.com.

[Scott Violet]

On Wed, Jul 1, 2015 at 5:32 PM, Dirk Pranke <dpr...@chromium.org> wrote:
> If we needed to, we could add the other mode to GN as well.
>
> Or, we could have a "hybrid" model where if the input list of targets
> contained groups, we would only include the targets in the group that
> are actually affected.
>
> I don't know that I have a strong leaning here myself.
>
> On Wed, Jul 1, 2015 at 1:48 PM, Scott Violet <s...@chromium.org> wrote:
>>
>> This is why analyze has the two outputs. For build only bots if you
>> compile build_targets that should have the same effect as building
>> all.
>
>
> You mean "same effect" in the sense that you'd rebuild all of the targets
> affected by the patch, right? I would expect building "all" to also cause
> you
> to rebuild targets that were out-of-date but otherwise unaffected by the
> patch
> (they are just out of date because the build was otherwise out-of-date).

Yes, your are correct.

-Scott