Re: AC_PATH_PROG ignores the variable

[Stepan Kasal]

Hello,

On Sat, Nov 06, 2004 at 11:13:17PM +0100, Alexandre Duret-Lutz wrote:
> Basically it says that when the output
> variable is an absolute path it overrides the output of the
> test, but when the output variable is not absolute it overrides
> the *input* of the test.  This I/O mix sounds awkward to me.

I'd say that if the variable looks like an absolute path, it overrides the
whole test, if not, it overrides only part of it.

If you want to be consistent, and your interpratation is that ``the
variable overrides the output of the test'', you should simply accept
the variables value, if it is set, without looking at it's value.

Please note that currently string ::/www overrides the test, while
../../bin/x doesn't.  (The former looks like a DOS path.)

> I'd rather keep the current semantic where setting an output
> variable can only affect the output: either the test uses the
> input of the maintainer, or it skips the test and uses the
> output supplied by the end user.

That sounds very reasonable.  But the border between the two alternatives
should be less obscure than it currently is.

My proposal was: "if it's not an absolute path, try to find the full path"
now I see it's more consistent to go along the lines of "if the variable
is set, take it as the output of the test".

With current code, the test is executed, even if the user tried to
override it.  It should issue a warning, at least.

> (For instance the suggested change will break Automake's
> AM_PATH_PYTHON.)

As you already know, I think that the code of AM_PATH_PYTHON is unfortunate.
By using this misfeature, it confuses the readers who could otherwise learn
from it how to write autoconf tests.

With kind regards,
        Stepan Kasal