[bug #31248] .SECONDEXPANSION: Inconsistent expansion of glob patterns in pattern rule dependency lines

[Paul D. Smith]

Follow-up Comment #1, bug #31248 (project make):

I've tested this with the latest GNU Make 4.4(.1).  I didn't really follow all
the discussion of previous behaviors, but here is how things currently behave,
which doesn't seem right to me:

Given the setup below, when run without second expansion:

$ make SEC=no
dir/foo.x from dir/foo.ybot

we can see that the directory prefix "dir/" is added to the pattern "%y*"
_before_ the glob is expanded, so that the expansion finds the pre-existing
dir/foo.ybot.

However if we enable second expansion:

$ make SEC=yes
dir/foo.ytop
dir/foo.x from dir/foo.ytop

we can see that here, the pattern is replaced giving "foo.y*" and the glob is
expanded before the directory prefix "dir/" is added, yielding foo.ytop which
is then changed to "dir/foo.ytop".

Clearly the fact that these work differently is wrong.

As for which should happen, my preference would probably be to change the
second expansion one to work like the no-second-expansion one, where first the
directory prefix is handled then the glob expansion is done afterward.  Adding
the directory prefix is part of expanding the pattern, and expanding the
pattern should happen before the glob match.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #31248] .SECONDEXPANSION: Inconsistent expansion of glob patterns in pattern rule dependency lines

[anonymous]

URL:
  http://savannah.gnu.org/bugs/?31248

 Summary: .SECONDEXPANSION: Inconsistent expansion of glob
patterns in pattern rule dependency lines
 Project: make
Submitted by: None
Submitted on: Mon 04 Oct 2010 10:43:47 PM UTC
Severity: 3 - Normal
  Item Group: Bug
  Status: None
 Privacy: Public
 Assigned to: None
 Open/Closed: Open
 Discussion Lock: Any
   Component Version: 3.82
Operating System: Any
   Fixed Release: None
   Triage Status: None

___

Details:

(The test case is at the end.)

In GNU Make 3.80, glob wildcards were not expanded in pattern dependency
lines.  In 3.81 they were expanded, but sometimes in the wrong directory.  In
3.82, we get the 3.81 behavior when second expansion is triggered, and
otherwise the 3.80 behavior.

Should glob expansion be suppressed for pattern dependency lines, as was
historically the case, even after second expansion?  If not, then it may be
best to prefix the stem directory before globbing, instead of after globbing.

Consider this makefile:

ifeq ($(SEC),yes)
.SECONDEXPANSION:
TRIGGER=$$(TRIGGER_SECOND_EXPANSION)
endif
all : dir/foo.x
%x : %y* $(TRIGGER) ; @echo '$@' from '$+'
foo.y% : ; @echo '$@'


(The TRIGGER stuff can be necessary to ensure that the second expansion
code path is followed.)  Consider also this directory tree:

.
./foo.ytop
./Makefile
./dir
./dir/foo.ybot


Linux GNU Make 3.82 and 3.80 with no command line arguments (the historical
behavior):

dir/foo.y*
dir/foo.x from dir/foo.y*


Linux GNU Make 3.82 with SEC=yes on the command line, and also 3.81
regardless of command line arguments:

dir/foo.ytop
dir/foo.x from dir/foo.ytop


Note that globbing matches foo.y* to foo.ytop in the top-level directory,
even though GNU Make then prefixes dir/.  If globbing is desired, then I
think it would make more sense to match dir/foo.y* against dir/foo.ybot.

(Thanks for reading this lengthy discussion.)

-- John Carey





___

Reply to this item at:

  http://savannah.gnu.org/bugs/?31248

___
  Message sent via/by Savannah
  http://savannah.gnu.org/


___
Bug-make mailing list
Bug-make@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-make