Re: XXX_CFLAGS

[Tom Tromey]

>>>>> "Ralf" == Ralf Corsepius <address@hidden> writes:

Ralf> * I would expect <program|library>_CFLAGS to be handled similiar
Ralf> to INCLUDES or DEFS (except for not being applied globally), but
Ralf> I do not expect it to rename files.

I think it is clear that if there is a conflict then renaming must be
done.

Ralf> * Similarily, I would expect hello_c_o_CFLAGS to be applied to the
Ralf> compilation of a single *.c/o's only.

I had considered that feature, but there didn't seem to be a very big
call for it -- most of the reports I had seen centered around the
inability to compile two programs in different ways.  Since there were
also design issues with this feature, I chose not to implement it at
the time.

Of course we can always add useful features, if there is time and
demand.

For instance, I think it would be logical to have a way to say
"programs A and B share these files (they are compiled the same way),
while these files must be compiled differently (and maybe renamed)".
I didn't think of that when I implemented this feature.

Ralf> * I don't see why using a single set of per-target CFLAGS causes
Ralf> the necessity of rename the files.

Other than occasionally leaving dead `.o' files around when you add
sources, I guess there's no problem.

Ralf> * Stylistically (A pure matter of personal taste), I am in favor
Ralf> of compiling into subdirectories (debug/profile etc) instead of
Ralf> renaming files.

This sounds reasonable.

Ralf> Furthermore, I am concerned that renaming the files will
Ralf> probably break custom suffix-rules and further custom rules
Ralf> applied somewhere.

In order to do per-exe flags, we have to generate explicit rules for
all such objects.  So suffix rules don't directly apply.

Ralf> I am not sure, but I suspect the naming not being safe against
Ralf> DOS/Win filenaming conventions (At the moment, I don't have
Ralf> access to such systems, so can't validate this statement).

I agree.  That is a problem.  That is why the _SHORTNAME feature was
introduced.


I agree there are problems with this approach.  And I think we should
work them out.  We can implement whatever we agree is reasonable.
However, I'd really prefer to get 1.5 out, even if it is less than
perfect.  It has waited far, far too long.  So my preference is not to
change these things now.  They are suboptimal, but don't appear fatal.

Tom