Mesa git testing experiences

[Schmaker]

Just a hint: It is helpful to refer to the exact version you are using, since "today's" and "last week's" or similar is not so precise. In a PPA version like 7.9.0~git20100819+gallium+r300.076c5387-0ubuntu0tormod the "076c5387" designates uniquely the git version (or commit ID).

Especially for regressions it is good to provide the last working tested git commit ID and the first broken one. Developers love commit ids

actually im not sure which one it was, thats why i posted date.

[nanonyme]

Speaking of functionality, I was somewhat impressed that r600g gave a pleasant (mostly) experience in openarena for me. This was with very high quality settings and resolution 1920x1280 and powersaving at maximum. Only really lagged once during a whole round. ^^

[rob2687]

I wanted to try git bisect earlier but something didn't seem to be happening right.

After checking out the master branch of mesa I did:

git bisect start
git bisect bad
git bisect good 15a3b42e135a3a2cb463ec3cff80a55dd8528051

This results in

Bisecting: 838 revisions left to test after this (roughly 10 steps)
[522de3f5ecbfe3f84e92ac03d1438a44eb1beae7] Close memory leaks from generate_constructor_intro

but commit id 522de3f5ecbfe3f84e92ac03d1438a44eb1beae7 != 15a3b42e135a3a2cb463ec3cff80a55dd8528051 ??? It bisects to a much earlier commit id. Is this how it's supposed to work?


Also I tried building it anyways using the usual ./configure --with-whatever-options that I use to build the r300g driver but it dies with

configure: WARNING: unrecognized options: --with-x, --disable-motif, --enable-xcb etc...

What am I doing wrong? What I am trying to do is use git bisect to find the revision where HoN started rendering funny by compiling r300g for each bisected revision until the most recent good commit. Is this right?

[tettamanti]

I wanted to try git bisect earlier but something didn't seem to be happening right.

After checking out the master branch of mesa I did:

git bisect start
git bisect bad
git bisect good 15a3b42e135a3a2cb463ec3cff80a55dd8528051

This results in

git bisect start
git bisect bad
git bisect good 15a3b42e135a3a2cb463ec3cff80a55dd8528051

but commit id 522de3f5ecbfe3f84e92ac03d1438a44eb1beae7 != 15a3b42e135a3a2cb463ec3cff80a55dd8528051 ??? It bisects to a much earlier commit id. Is this how it's supposed to work?

Yes, it's a bisection Given two known points (in this case the ones that you marked good and bad) it picks a commit near the middle of the history between the two points. Once you've tested the proposed commit you can exclude one of the halves of the history and redo the bisection on the other half.

configure: WARNING: unrecognized options: --with-x, --disable-motif, --enable-xcb etc...

Hum, try and regenerate the configure script...

[nanonyme]

motif is disabled by default so --disable-motif doesn't exist

[rob2687]

It goes to a commit much earlier than the id for the commit marked as good though.

Entered this one (2010-08-17)

http://cgit.freedesktop.org/mesa/mesa/commit/?id=15a3b42e135a3a2cb463ec3cff80a55dd8528051

but it goes to this one (2010-06-24)

http://cgit.freedesktop.org/mesa/mesa/commit/?id=522de3f5ecbfe3f84e92ac03d1438a44eb1beae7

That doesn't seem to make much sense to me.


About the ./configure thing it rejects every option I pass it aside from --prefix=.

These are the same options I use to regularly compile the most recent master/head or whatever.

[tormod]

It goes to a commit much earlier than the id for the commit marked as good though.

Did you tell git which commit is bad?

[whizse]

Did you tell git which commit is bad?

Just using "git bisect bad" should mark the current version as bad.

[whizse]

It goes to a commit much earlier than the id for the commit marked as good though.

Entered this one (2010-08-17)

http://cgit.freedesktop.org/mesa/mesa/commit/?id=15a3b42e135a3a2cb463ec3cff80a55dd8528051

but it goes to this one (2010-06-24)

http://cgit.freedesktop.org/mesa/mesa/commit/?id=522de3f5ecbfe3f84e92ac03d1438a44eb1beae7

That doesn't seem to make much sense to me.

I _think_ this is because the commit in question was developed on the glsl2 branch, and only recently merged to master. But I'm far from an expert on git...

[rob2687]

Well bisecting from anything after the glsl2 branch was merged works fine. Taking a commit from before it does the above.