In response to (1):

This makes no difference. I'm not sure if we're on the same
planet. I'm not quite sure what the patch was meant to
achieve (and thus I don't understand what I was supposed
to do to take advantage of it once configure was
regenerated). I think the loop that fails to find libpng
is indeed the one you've provided the patch for, so you
and I are possibly within the same universe.

In response to (2):

> Since you obviated a system immanent feature...

Hey, I'm really confused now. I'm not at all sure what
nuance you're implying with those words. I really
don't understand why you said it at all. Can I try
saying this to you:

/usr/local/include/libpng/png.h (for both arch)
/usr/local/include/libpng/pngconf.h (for both arch)
/usr/local/lib/libpng12.so (32-bit)
/usr/local/lib/sparcv9/libpng12.so (64-bit)

PHP needs to use the files in /usr/local/include/libpng
and /usr/local/lib/sparcv9. The library path is already
known by the compiler, linker, and loader.

If you want support your environment we would have to change all configure files. We would have to change all
lines of the form .../lib/... with ../$LIB_DIR/... and
add some configure magic to determine what $LIB_DIR should be (in your case it would be sparcv9).

We can add a new macro to the configure, which is used 
always for the direct search of a library files. Now the list of common paths are:

/usr/local/lib /usr/lib

With 64bit linux distros:

/usr/lib/lib64/

(not sure if if e.g. /usr/local/lib64 can exist too?)

Sparc64:

*/lib/sparcv9/

I'll look into adding the macro to make the configure
be a bit friendlier. :)

> I'll look into adding the macro to make the configure
> be a bit friendlier.

Thank you for looking into this and suggesting something that would let
users to work around the brain-dead ./configure design. It will be
helpful if the problem and its solution are obvious to people (e.g. if
it is possible to generate good error messages).

At the time of my original submission, PHP was just using a stupid
library detection method. This was with regard to PNG and JPEG support
-- which almost all other software (maybe not xpdf -- I can't recall)
seems to be able to find by itself. Presumably this is because such
software uses the linker to carry out the tests. Perhaps PHP has some
new requirement that prevents it from carrying out a conventional test?

It would have made more sense if PHP only fell back to its brute-force
file matching if a linker test failed. In fact, I think my solution to
PHP's behaviour was to delete out a few 'exit' statements -- no actual
practical problem existed. I think that it got worse with 4.3.2 because
the faulty configuration motif occurred multiple times within
./configure.

> Sparc64:
>
> */lib/sparcv9/

PHP seems to be in an exclusive club of software that requires this
extensive hand holding.

Problems that are not solved by this workaround:
 - PHP needs to be taught about every different distro,
 - PHP needs to be told about each different site.

For example, Solaris on UltraSPARC supports multiple ABIs (and
presumably Mac OS X and Linux now do, too). A particular *site* may
choose to compile PHP for a 32-bit OR a 64-bit ABI (or both, though that
is only likely to occur when there are site-specific constraints that
require it). Thus PHP can't be shipped with this prior knowledge -- it
needs to be told on a site-by-site basis. In my scope of activities, the
only other software requiring this much ABI knowledge is that which uses
assembly code (e.g. OpenSSL).

Another 64bit issue found in bug #24950 (oracle)

Use --with-libdir, we hold hands here. ;)