[bug #18622] regression cannot turn off implicit rules for %.c and %.tex in 3.81

[Paul D. Smith]

Follow-up Comment #2, bug #18622 (project make):

Ouch.  This is a result of the secondary expansion feature.  The problem is
that we defer the tokenization of dependencies until the snap_deps() step,
when all the makefiles have been read in.  This means that when we try to
override the pattern rule and compare the dependencies, they don't match
because our "internal" dependencies are two separate strings "%.w" and "%.ch"
(for example) but the overriding dependencies are (at this point) a single
string "%.w %.ch".  We don't split them until later.

There are actually a number of problems with the current secondary expansion
feature, not least of which is that it wastes memory; I have a half-completed
workspace that is changed to always split dependencies right away UNLESS we
can show that secondary expansion is a possibility (that is that the
.SECONDARYEXPANSION special target is present, AND that the prerequisite list
contains expandable elements---contains one or more "$").

Once that change is completed, it will fix this bug as well.

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?18622>

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