This is quite fragile, but it works for now, unlike the python
version.
In general it seems conf.py is not intended to be evaluated outside of
sphinx, as it assumes certain global names (in particular "tags") are
defined.
Check that copyright year will be current year in generated documentation.
Checking is done my matching that copyright line contains current year
as a substring which is good enough "approximation" in this context.
- Make lib/notmuch.h the canonical location for the library versioning
information.
- Since the release-check should never fail now, remove it to reduce
complexity.
- Make the version numbers in notmuch.h consistent with the (now
deleted) ones in lib/Makefile.local
Before release check that there are no uncommitted changes and
that there are no files in working directory that possibly should
have been added to the repository.
Amended by db: remove --ignored, since that seems like too much
trouble.
As discussed in
id:8cc9dd580ad672527e12f43706f9803b2c8e99d8.1405220724.git.wking@tremily.us,
execfile is unavailable in python3.
The approach of this commit avoids modifying the python module path,
which is arguably preferable since it avoids potentially accidentally
importing a module from the wrong place.
Manual pages are now generated and during the generation the version
string is read from `version` file, so this (currently failing) test
checking manual page versions can be removed.
While at it, changed the case pattern *[^0-9.]*
to its portable alternative *[!0-9.]*
NOTMUCH_VERSION_* macros in lib/notmuch.h are replaced with
LIBNOTMUCH_VERSION_* macros. Check that the values of those
match the LIBNOTMUCH_*_VERSION values in lib/Makefile.local.
New defines NOTMUCH_MAJOR_VERSION, NOTMUCH_MINOR_VERSION and
NOTMUCH_MICRO_VERSION were added to lib/notmuch.h.
Check that these match the current value defined in ./version.
Check that the underlining '===...' for first (header) line in NEWS
file is of the same length as the header text and it is all '=':s.
-- extra execs removed by db.
Version string has strict format requirements in release-check.sh:
only numbers and periods (in sane order) are accepted.
Mismatch there used to halt further execution.
In this case, checking versions like '*~rc1' for (more) problems
was not possible.
This 'fatal error' is now changed buffered error message like in
following tests, and is displayed at the end of execution.
Currently Makefile.local contains some machine executable release
checking functionality. This is unnecessarily complex way to do it:
Multiline script functionality is hard to embed -- from Makefile point
of view there is just one line split using backslashes and every line
ends with ';'. It is hard to maintain such "script" when it gets longer.
The embedded script does not fail as robust as separate script; set -eu
could be added to get same level of robustness -- but the provided
Bourne Again Shell (bash) script exceeds this with 'set -o pipefail',
making the script to fail when any of the commands in pipeline fails
(and not just the last one).
Checking for release is done very seldom compared to all other use;
The whole Makefile.local gets simpler and easier to grasp when most
release checking targets are removed.
When release checking is done, the steps are executed sequentially;
nothing is allowed to be skipped due to some satisfied dependency.
