❌ Build nut 2.8.2.2817-master failed (commit d9c4063805 by @sshambar)

Great catch, reasonable analysis and fix - thanks!

How did you stumble on this? On many systems, make distcheck{,-light} did pass - although this is a test to build it in same conditions as the original creation of the archive, so checking in another bare-bone system would have been more reliable.

I did have concerns about better checks of the archive validity, posted at #2829... I suppose your PR brings the solution a step closer :)

✅ Build nut 2.8.2.2818-master completed (commit 8ce102b6c2 by @jimklimov)

How did you stumble on this? On many systems, make distcheck{,-light} did pass - although this is a test to

I was trying to fix the Fedora RPM to build HEAD, and make distclean; make dist to make a source-tar always failed with
cp: cannot stat './nut.conf.5': No such file or directory
but running dist check again worked... traced the problem to the missing suffix rule. (I think distcheck worked because it builds html docs, which masks the man the problem).

Of course, then trying to build the RPM failed to install the man pages (because the install-man: rule was empty), and figured how the man custom-section change caused the automake issues... testing my fix exposed the missing index.txt dist file...

I found a couple other issues updating the RPM, but I figured fixing the build was best done first :)

I think the changes about prepped rules broke the build, at least for Sun make.

Your new ruleset seems factually wrong in any case (maybe that is what upsets the build):

+.txt.txt-prepped:
+       rm -f "$(abs_top_builddir)/docs/man/.prep-src-docs"*
+       $(MAKE) prep-src-docs

The .prep-src-docs is a single flag for dependency tracking (newer than all of its prerequisites) so you remove it every time you prep another text file, and then call make prep-src-docs which processes all PREP_SRC files (and makes their individual *-prepped contents). I suppose that new code has potential to overlap, deleting the flag and re-prepping all the files for each file.

Also, calling autotools make is more complicated than that (see other cases with AM_MAKEFLAGS, + for job server cooperation, etc.

Currently it seems to end up with no *-prepped files in distcheck work area, and fails to generate mans/htmls from them.

Also, the .NOTINTERMEDIATE keyword seems to be GNU Make specific, not an automake scripting trick, right?.. In fact, IIRC the percent-rules like %.txt-prepped are also a GNU make extension, not a standard thing.

Was this bit critical for your fixes? I'll try to put back the previous code for these, to see if it helps.

The problem I hit with .prep-src-docs as it was used is that it "satisfied" it's first target (nut.conf.5 in my case), as nut.conf.5 didn't have any explicit recipes... the correct solution would be to have the .txt.txt-prepped rule actually build individual files (so .prep-src-docs isn't needed at all), but you have some fancy business in the .prep-src-docs recipe that I didn't want to mess with...

Reverted... nut.conf.5 will still fail on make distclean; make dist, so the logic for the flag needs to be handled some other way (my recursive make is just what I've seen used for "meta-flags" that handle multiple targets as a batch... but yeah it's probably not universally portable :)

I guess .NOTINTERMEDIATE probably is a GNU thing, would .PRECIOUS be portable?

Actually nut.conf.{5,html} seems to have been a tip of the iceberg, just first in list. A make distcheck -kshowed all (or a lot of) man pages to fail before I reverted that prep part of the fix :\

fail on make distclean; make dist

Aren't you supposed to re-configure in between?

There's also a number of files generated by autogen.sh as templates for configure (so they are EXTRA_DIST'ed but are not in git, and I had a case with files generated from those templates going AWOL at make clean. Possibly the {,DIST,MAINTAINER,...}CLEANFILES lists need a revision if such scenarios fail (helped in that case)...

Seems the main trick in this regard was about BUILT_SOURCES (fixed in newest commit above). And having the tools to build man pages (when building from git sources), or having them originally (from release/dist tarball).

❌ Build nut 2.8.2.2821-master failed (commit 5ff494b65a by @jimklimov)