Re: possibly undefined macro: dnl

[Akim Demaille]

>>>>> "adl" == Alexandre Duret-Lutz <address@hidden> writes:

>>>> "Akim" == Akim Demaille <address@hidden> writes:
adl> [...]

Akim> * Output the actual line where the culprit was found, i.e., not
Akim> the _input_ file, but the _output_ file.

adl> This would be more helpful, methinks.

Agreed.

Akim> * Look for the culprit token in all the sources in reversed
Akim> order and output it with an error message that clearly
Akim> demonstrates that it's only a guess.

adl> I understand it's better to grep all the sources, and change the
adl> error message.  However I fail to see how ordering can help,
adl> because you don't know which occurence of the macro is guilty.

Sure.  But the hypothesis is `the first files have been more tested
and are less likely to be guilty'.

For instance, it makes no sense to look for an unexpanded AC_TRY_RUN
in autoconf.m4, or an AT_CHECK in autotest/general.m4.  The idea is
user files are last, and user files are more suspicious than lib
files.

So since we should not grep all the sources (think of this dnl for
instance:

~/src/ace % grep -c dnl <lib/{m4sugar,autoconf}/*.m4             nostromo 12:17
633

Gosh, I really *love* zsh!!!), we should output just one, and the last
one I think.