implementation now has its own directory under sapi/, just like
extensions have theirs under ext/. To make the final targets appear
in the main dir, the top-level Makefile includes sapi/NN/Makefile.inc
from the selected sapi backend. This is a plan Makefile stub without
any autoconf substitutions. Each SAPI backend also has its own
config.m4 like extensions (read at the end of diversion 2) and
config.h.stub files.
Each SAPI backend has to contain:
config.m4: just like for extensions, this file contains
autoconf/automake directives that end up in the configure script. The
only difference is that the sapi config.m4 files are read in diversion
(output block) 2 instead of 3. The sapi config.m4 files should set
two variables: PHP_SAPI (which sapi backend to choose) and SAPI_TARGET
(the name of the resulting library or program, previously BINNAME).
If they are not specified, they will default to "cgi" and "php",
respectively.
Makefile.inc: has to exist, has to define "INSTALL_IT" to the command
used to install the final target (or ":" for no operation). It also
has to define a plain Makefile rule (without autoconf substitutions)
to build $(SAPI_TARGET)
Makefile.am: just what you think. Make sure your target is called
"libphpsapi_NNN.a", where NNN is the value of PHP_SAPI.
Some testing and fixing probably remains. To make everything hang
together, I've done some ugly tricks that I can imagine causing some
problems. I've built and run the CGI version and built the Apache
DSO.
possible. We still need to set our connection status bit field so a
shutdown function can figure out if a script was terminated due to an
abort, timeout or finished normally.
yet. That is coming next. Have also only done Apache and CGI so far.
Will have to crack open my ISAPI book to do that one.
Also changed the SAPI output functions to return an int. We'll check
the connection status inside each one, but we might need the return
code at some other level and I don't see a good reason for just tossing
these return codes.
Thies C. Arntzen