waf needs some tlc

./waf distclean doesn't remove $BUILD/ntpd/version.h ./waf distclean doesn't remove $BUILD/ntpclients/ntp

./waf build --notests builds lots of test stuff [125/191] Compiling tests/unity/unity.c [126/191] Compiling tests/unity/unity_fixture.c [127/191] Compiling tests/libntp/a_md5encrypt.c [128/191] Compiling tests/libntp/authkeys.c [129/191] Compiling tests/libntp/calendar.c [130/191] Compiling tests/libntp/clocktime.c ... [158/191] Compiling tests/common/tests_main.c [158/191] Compiling tests/common/caltime.c [160/191] Compiling tests/common/sockaddrtest.c [161/191] Compiling tests/common/file_handling.c

./waf build --notests --disable-manpage builds lots of man pages. Do I have to put it on the configure step? If so, it should barf on the build line.

I don't have a handy example, but my scripts don't build the pythin .pyc/pyo stuff until I run the check script. I expect them to get built by the build step.

We no longer have a waf expert and yes, the recipe is difficult to modify. This is a known problem for which I don't have a solution.

--disable-manpage does need to be done at the configure step. I don't know any way to intercept or validate -- options at build time short of modifying waf.

I can't reproduce failure to build the Python stuff. Happens for me in the build step after either clean our distclean.

waf distclean is supposed to remove the entire build directory, and iy did when I tried it. Did you mean to reference the source directory?

--disable-manpage does need to be done at the configure step. I don't know any way to intercept or validate -- options at build time short of modifying waf.

Is that true for all options? Things like --notests or --disable-manpage make sense at build time. I'm happy to provide them at configure time. I'm just trying to understand things.

waf distclean is supposed to remove the entire build directory, and iy did when I tried it. Did you mean to reference the source directory?

Yes. Sigh. Sorry for the noise.

Yes, it does remove the build directory.

[murray@hgm raw]$ ./waf distclean 'distclean' finished successfully (0.018s) [murray@hgm raw]$ ll ntpd/version.h ntpclients/ntp lrwxrwxrwx 1 murray murray 19 Dec 18 20:24 ntpclients/ntp -> ../build/main/pylib -rw-rw-r-- 1 murray murray 704 Dec 23 01:17 ntpd/version.h [murray@hgm raw]$

waf build generates:

__init__.pyc
statfiles.pyc
packet.pyc
util.pyc
statfiles.pyo
__init__.pyo
util.pyo
packet.pyo
ntpc.so

waf check generates:

magic.pyc
control.pyc
version.pyc
magic.pyo
control.pyo
version.pyo

waf build should have done all of that. ctx.add_group() is called after the ctx() calls to build control, magic, and version.

I see this behavior with the 0.9.6 release tarball, but not with git HEAD.

I see this behavior with the 0.9.6 release tarball, but not with git HEAD.

Did you try it on a clean clone?

Your "this" is ambiguous and/or I lumped several quirks into one issue. There are interactions between not cleaning fully and not being able to reproduce rebuilding at check time rather than build time because leftover files don't need to get rebuilt.

I used an existing clone, but I removed the entire build/ directory manually. Was that not sufficient to clean things out?

I used an existing clone, but I removed the entire build/ directory manually. Was that not sufficient to clean things out?

Maybe. Maybe not.

Some of the derived files were put into the source directories. For example, there is a ntpd/version.h

Eric may have fixed some of that.

Best to try it with a fresh clone. I suggest cloning to a directory that isn't used for anything but making local clones. Then clone again from there to run an experiment. Local clones are super fast. Throw away the experimental directory and clone again as needed.

added build label

mentioned in commit 218ee2ae

State of this issue:

waf distclean still doesn't remove $SRC/ntpd/version.h or ntpclients/ntp. This can't be fixed due to a limitation of waf - distclean has an opaque execution path that's completely separate from clean. I'm going to have to mark that "won't fix" and nag ita about the deficiency in waf.

The complaint about --notests --disable-manpage was based on a nisunderstanding. Those options need to be given at configure time.

Following the upgrade to waf 1.9.13, the failure to build python stuff until waf check in a fresh clone is gone. Instead, it now is done at build time, but if you do waf check it gets redone. I think this was a manifestation of some bug in thw waf inference engine, triggered by unusual productions, that has been part fixed. We can live with the current behavior.

Accordingly, I'm closing this bug. If any of the individual subissues need to be revived, please do that in a separate issue per problem.

closed