* Eric Blake wrote on Mon, Jan 28, 2008 at 09:50:49PM CET: > According to Ralf Wildenhues on 1/28/2008 1:37 PM: > | Isn't this an API change? It looks to me the at* files are now > | interpreted after the command line (untested). That seems wrong: > | the command line should override atlocal, if I do > | ./testsuite PERL=/some/other/perl > > Ouch. I think you're right. The problem which drove me to move this > snippet was that unless atconfig lives in the same directory where you > invoke testsuite, it does not get sourced, which is not what we want > either (in the case of m4, it meant that running tests/testsuite did not > discover the default AUTOTEST_PATH=tests in $(builddir)/tests/atconfig, > and therefore, the test was being run on the installed m4 rather than the > just-built m4).
Sure. I see the problem that you are trying to solve. > I guess the correct solution is to look for atconfig in the current > directory (as was the case before my regression) as well as in dirname > "$at_myself" (which may be the same directory), before declaring that it > can't be found. I'll work on a patch along those lines. Erm, `dirname "$at_myself"' is usually (exception: the Autoconf package itself) in the source tree, while atconfig is usually in the build tree. I don't see how to solve this without parsing the argument list twice (scanning for -C first), or somehow deferring the setting of the variables which are passed on the command line. Cheers, Ralf
