Re: non-recursive makefiles

[Assaf Gordon]

Hello,

Attached an improved patch.

> On Jun 10, 2016, at 01:27, Assaf Gordon <address@hidden> wrote:
> 
> Some open issues:
> 2. 'make distcheck' fails on Mac. The error is:
>     /usr/local/bin/texi2dvi: etex exited with bad status, quitting.
>   (perhaps this is unrelated).

This turns out to related to the build, but more complicated:
Seems like the current 'doc/Makefile.am' (and my './doc/local.mk') have some 
inconsistencies that will be better handled in a dedicated patch:
1. the resulting file 'doc/sed.texi' is auto-generated from 'doc/sed-in.texi' , 
yet it is also checked into git (and same for 'doc/s-texi' ).
2. the auto generated files are written to '$srcdir', not '$builddir' (this 
also applies to the man page).

The specific failure with 'make distcheck' was due to running 'make 
doc/sed.dvi' from an out-of-tree build,
where the source 'doc/sed.texi' is not in the current or parent directories.

The attached patch does not fix any of these issues.

> 3. The non-shell tests (those based on ./testsuite/runtest and 
> ./testsuite/Makefile.tests)
>  save their log/trs files in the top directory instead of ./testsuite.

Fixed by adjusting the path of the tests.
But this introduced a minor issue:
'testsuite/get-mb-cur-max' (which is a utility program)
is now treated as a stand-alone test. An ugly hack is to skip it
in 'testsuite/runtest' - but there should probably be a nicer solution.


> 4. running 'make' when ./sed/sed does not exist emit errors such as:
>    $ rm -f sed/sed
>    $ make
>    make: ./sed/sed: Command not found

Fixed:
The old 'doc/Makefile.am' redefined the SED variable (to pass it to help2man).
This overrode the system-wide SED set by automake/autotools.
Renamed to 'SEDBIN' in 'doc/local.mk'.


Comments welcomed, 
 - assaf

0001-build-switch-to-non-recursive-makefile.patch.xz
Description: Binary data