IMHO everything is fine and we don't have to do anything ;-)

Antoine:

I've noticed that Travis-CI (sometimes?) compiles CPython twice.

"make regen-all" doesn't compiles Python:

• it compiles Parser/pgen
• it compiles Programs/_freeze_importlib
• it runs many commands to regenerate generated files

2017-07-24 9:37 GMT+02:00 Serhiy Storchaka <storchaka@gmail.com>:

make regen-all touches header files which are dependencies for all binaries. I suggest to run make regen-all before make.

Zachary Ware explained me once that "make regen-all" should be run after "make", but I don't recall why :-)

Travis CI makes sure that "make regen-all" doesn't modify generated files (or the build fails), so if we would run "make" a second time after "make regen-all", it would be supposed to do nothing.

Victor

I read again the read and I misunderstood it. I didn't notice that "make buildbottest" recompiles Python. Ok, I confirm the bug.

I see two options:

• move the "make regen-all" check *before* "make"
• move the "make regen-all" check in a different Travis CI. If we move the test in a GCC job, it would allow to check regen and GCC warnings in the same job.

Victor

2017-07-24 12:16 GMT+02:00 Antoine Pitrou <antoine@python.org>:

Le 24/07/2017 à 10:55, Victor Stinner a écrit :

...

IMHO everything is fine and we don't have to do anything ;-)

What do you mean? The fact that all the CPython source code (including C extensions) is compiled twice looks suboptimal to me. We could probably win a minute or two on Travis-CI build times.

Regards

Antoine.

---

python-committers mailing list python-committers@python.org https://mail.python.org/mailman/listinfo/python-committers Code of Conduct: https://www.python.org/psf/codeofconduct/

Technically, "make regen-all" doesn't use the freshly built Python. It uses PYTHON_FOR_REGEN which is usually "python3".

Victor

2017-07-24 15:48 GMT+02:00 Nick Coghlan <ncoghlan@gmail.com>:

On 24 July 2017 at 18:55, Victor Stinner <victor.stinner@gmail.com> wrote:

...

2017-07-24 9:37 GMT+02:00 Serhiy Storchaka <storchaka@gmail.com>:

...

make regen-all touches header files which are dependencies for all binaries. I suggest to run make regen-all before make.

Zachary Ware explained me once that "make regen-all" should be run after "make", but I don't recall why :-)

Some of the generators (including Argument Clinic) are themselves written in Python, so building the checked in version first is the only way to be 100% sure you have a compatible version available.

As Serhiy notes, the robust fix is to make sure the generators leave the file modification times unchanged if they don't actually change anything, either by working entirely in memory and only writing the result back out if it changed, or by generating out-of-place and then doing either a rename (if the result changed), or deleting the new one (if it is the same as the original).

Cheers, Nick.

-- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia