Re: Slow test execution for usr.bin/make

[Robert Elz <kre%munnari.OZ.AU@localhost>]

martin%duskware.de@localhost said:
  | Right, but the external tr process will not be run for each of them,
  | so there will be a serious speedup (and we have other test programs
  | with quite high number of test cases).

I am currently testing a (fairly simple) mod to tests/usr.bin/make/t_make.sh
which will remove that command substitution running tr with shell code.

Normally, replacing tr with shell scripting would be a poor choice, but
here the speedup is likely to be considerable - the actual translation of
'-' into '_' will certainly be slower than tr could do it, but we avoid
the cost of 2 forks (one for the cmdsub which is a full fork,  and one for
tr which at least would just be a vfork) and an exec of tr for every test
case.

The code is written and tested, I am just doing a build of HEAD (actually
what was HEAD a while ago now, but that shouldn't matter) to verify that
the make ATF tests still work properly with this change.

Adding extra mechanism to construct t_make.sh at build time (lumping
in all of the tests then - and making sure that the dependency list for
doing that is correct and remains so) might not be required.

kre