Re: RFA: automatic dependency tracking

[Tom Tromey <tromey@redhat.com>]

>>>>> "Ralf" == Ralf Wildenhues <Ralf.Wildenhues@gmx.de> writes:

Ralf> [ChangeLog]
>> (cli-decode.o, cli-dump.o, cli-interp.o, cli-logging.o,
>> cli-script.o, cli-setshow.o, cli-utils.o): Likewise.

Ralf> Missing cli-cmds.o.

Tough crowd :)

Ralf> s/by the GNUMakefile/by GNU make-specific code below/

Yeah, I thought I'd gotten that one, but I guess not.  Thanks.

Ralf> BSD make will complain about a rule command
Ralf> 	$(POSTCOMPILE)
Ralf> when the macro expands to empty; let's set it to '@true' instead.

I don't mind either way.  This must be version-specific, though, since
this patch was tested on BSD, specifically FreeBSD 6.0-Release:

    http://sourceware.org/ml/gdb-patches/2008-07/msg00586.html

What version are you using?

Ralf> BSD make barfs over $< outside of inference rules.

Ditto.

>> # gdb/cli/ dependencies
>> #
>> # Need to explicitly specify the compile rule as make will do nothing
>> # or try to compile the object file into the sub-directory.

Ralf> Have you thought of simply compiling the object file in the sub
Ralf> directory?  That's not a bad option; and the makefile has already been
Ralf> assuming that the compiler understands '-c -o' anyway.

That comment predates my patch.  I just preserved it.

FWIW I did not want to change where the build puts objects; I didn't
see a particular benefit to doing this.

>> +@GMAKE_TRUE@$(all_object_files): | $(generated_files)

Ralf> This order-only dependency requires make 3.80 or newer.

Ralf> Not sure if you want to change this, but I guess the documentation
Ralf> should otherwise be updated to reflect that GNU make < 3.80 can't be
Ralf> used.  (Or you disable GMAKE_TRUE for that...)

3.80 was released in October 2002.  So, I am not really worried about
this myself.  I didn't see any docs to update.

We could dynamically detect 3.80 by looking at .VARIABLES.
(.VARIABLES was introduced in 3.80.  .FEATURES wasn't introduced until
3.81, so we can't use that.)

In the case where make is too old, we could emit a warning and not
enable the dependency tracking, like:

ifndef .VARIABLES
$(warning GNU make 3.80 or later is required for automatic dependency tracking)
else
... existing code
endif

(All protected with @GMAKE_TRUE@ of course.)

What do you think of that?

Of course, then we would need someone to test this with 3.79.

>> +dnl For dependency tracking macros.
>> +sinclude([../config/depstand.m4])

Ralf> Why sinclude?

That is what the rest of the file does.  In general I try to submit
patches which match the prevailing style.  If a cleanup is really
needed, then I send that as a separate patch.

Ralf> Adding the patch below on top of yours should fix all these issues.

Could you just send the full, updated patch?
Thanks.

Tom