751076 - pymake returns nothing for $< and $^ in some cases

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Mike Hommey [:glandium]]

Example of a failing Makefile:

FOO = foo.o
BAR = bar.o

$(FOO): %.o: %.c
$(BAR): %.o: %.cpp

$(FOO): a.h

$(FOO) $(BAR):
	echo $<
	echo $^

Comment 1

[Joey Armstrong [:joey]]

Also adding a check/unit test to verify functionality when pymake is changed would be helpful going forward.

Run (g)?make and pymake on the test makefile, report values generated by all automatic variables then strcmp the two sets for equality.

Comment 2

[(not currently active) Ted Mielczarek]

Conveniently pymake has a rather thorough testsuite that runs tests in both gmake and pymake:
http://hg.mozilla.org/users/bsmedberg_mozilla.com/pymake/file/9fb06df46292/tests

Comment 4

[Gregory Szorc [:gps]]

Attachment: Merge prerequisites for targets with multiple rules, v1

Assigning to glandium only because he filed the bug.

The pymake tests pass - ship it!

Try at https://tbpl.mozilla.org/?tree=Try&rev=23e4fdc98ae1

Comment 5

[Gregory Szorc [:gps]]

And this broke the build: https://tbpl.mozilla.org/php/getParsedLog.php?id=22469633&tree=Try&full=1#error0

Comment 6

[Mike Hommey [:glandium]]

(In reply to Gregory Szorc [:gps] from comment #5)
> And this broke the build:
> https://tbpl.mozilla.org/php/getParsedLog.
> php?id=22469633&tree=Try&full=1#error0

Dependencies are probably not handled correctly.

Comment 7

[Michael Shal [:mshal]]

Attachment: Fix $< and $^ for pymake

I can't access the previous tbpl results so I don't know for certain, but the original patch was failing locally for me because $< was being set incorrectly. In gmake, if we have:

baz: baz.in2

baz: baz.in1
     Use $<

Then it treats baz.in1 as the first prereq for $< purposes, even though baz.in2 is technically "first". If the rule with $< has prereqs, it uses the first prereq in the rule itself. However if it has no prereqs, then it will use the first prereq available in any prior rules.

I've updated the patch to match gmake here, and included this in the test case. This builds locally in pymake, and Windows try results are here:

https://tbpl.mozilla.org/?tree=Try&rev=33487368330a

Not sure if the oranges are related or not though.

Comment 8

[Gregory Szorc [:gps]]

The Try push looks good. Just a JS JIT test failure.

mshal: Thank you very much for fixing this. This bug is likely responsible for a bunch of required clobbers. It also caused much grief as I was implementing non-recursive XPIDL generation. This will make writing nonrecursive make files much easier.

Please don't forget to double land pymake patches into https://hg.mozilla.org/users/bsmedberg_mozilla.com/pymake/

Comment 9

[Michael Shal [:mshal]]

inbound: https://hg.mozilla.org/integration/mozilla-inbound/rev/1ad29e4c3564

Also pushed into the pymake repo.

Comment 10

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/1ad29e4c3564