Re: spurious install of ${prefix}/include/gps.h

[Greg Troxel]

Bisecting turned out cleaner than I expected, with no confounding
issues.  Every run either had a clean build or a clean build followed by
the spurious install attempt.

bisect tags things as:

  commit 2639b9826456679e9c89e22aebdbe21ee9adde63 (refs/bisect/bad)
  Author: Gary E. Miller <gem@rellim.com>
  Date:   Tue Mar 9 14:52:08 2021 -0800

      www/gpsd-client-example-code.adoc: Expand the example.

  commit 718492526ed918d11b3972d834080bc954449924 (HEAD, 
refs/bisect/good-718492526ed918d11b3972d834080bc954449924)
  Author: Gary E. Miller <gem@rellim.com>
  Date:   Mon Mar 8 19:20:32 2021 -0800

and looking at the diff, this is mysterious as the changes seem
obviously reasonable.

I built 2639b9826456679e9c89e22aebdbe21ee9adde63 but with the 1-line
change to Sconscript adding www/example1.c to htmlpages reverted.  That
did not exhibit the bug.

Next I built master, with the example1.c line removed.  That also built.

Then, I did a round trip through pkgsrc, doing a build and then scons
dist, and then using that tarball as the upstream release in pkgsrc,
building a package and installing it.


So I'm not sure how to fix, but I have figured out the change that
causes the bug.

I am guessing that scons is looking at the .c file that is copied to
html and deciding that because
  1) it includes <gps.h> and
  2) -I/foo/include is in CPPFLAGS while includedir=/foo/include that it
      must somehow install gps.h.

This doesn't make sense because

  1) the same rule should have fired with includedir=/usr/include and
     the implicit/default inclusion of /usr/include as -I, and
  2) The install is done after building, not before
  3) files copied to html are not compiled

I wonder if renaming example1.c to example1.c.example would help?


I'll do a live test with ublox later today.

Thanks to all for hints and testing.

signature.asc
Description: PGP signature