Re: shishi-0.0.12: "missing" script fails with help2man

[Alexandre Duret-Lutz]

>>> "Simon" == Simon Josefsson <address@hidden> writes:

 Simon> Adrian Bunk <address@hidden> writes:
 >> Hi,
 >> 
 >> I'm not sure whether the following build error is a bug in shishi or in
 >> the "missing" script:
 Simon> ...
 >> /bin/sh /aux/adrian/build/shishi-0.0.12/missing --run help2man \
 >> --name="kerberos 5 database interface" \
 >> --info-page=shishi \
 >> --output=shisa.1 ../src/shisa
 >> Unknown option: i
 Simon> ...
 >> shishi-0.0.12 uses the "missing" script from automake 1.8.
 >> 
 >> GNU help2man 1.24 is installed.

 Simon> Hello, and thanks for the report.  I'm using help2man 1.33, and most
 Simon> likely the cause for the error is that 1.24 is too old.

 Simon> But the consequence of the problem should, IMHO, be
 Simon> mitigated by automake/missing -- the man page is after
 Simon> all included in the shishi tar.gz.

If the problem occurred during a user build of shishi (by user
build I mean that no distributed files has been altered) then
the Makefile probably needs to be fixed.  Indeed, since the man
page is distributed, Makefile rules should be arranged so that
it is not rebuilt during a user build.  This kind of error is
usually caught by `make distcheck' and is discussed in the FAQ.

If the problem occurred during after some modification of the
distributed files, then obviously the above paragraph does not
apply.  It's just unclear to me which case we are in.

 Simon> So, even if help2man fails, the build should not abort,
 Simon> since there is no fatal error.  (Something must
 Simon> guarantee that the man page is not overwritten by a
 Simon> broken help2man, though.)

The problem is that _you_ know the error is just a version
incompatibility that could be ignored by reusing the existing
file, but missing has no way to know that: it just sees that
help2man exists and exited badly.  After all, missing does just
its job: emulating tools when they are missing, which is not the
case here.

One can legitimately argue that if Foo 4.2 is expected but only
Foo 2.0 exists, then Foo should not be considered present.  This
old problem is also mentioned in the FAQ (see the
maintainer-mode node), and caused some confusion during the
switch from autoconf 2.13 to 2.50...  One solution, introduced
in recent Autoconf and Automake versions is the use of a special
exit code to denote version mismatches (see NEWS from Automake
1.8), but this can apply only when there is a way to request a
minimum version.
-- 
Alexandre Duret-Lutz