Test fixes for DJGPP

[Tim Van Holder]

Tried the testsuite for the first time in a long time (despite
the many test files, it actually runs considerably faster than
autoconf's test suite - phew).
One general 'make check' nit - it would be nice if the list of
unexpected failures and successes was displayed at the end, for
those cases where the output of make check wasn't redirected
(otherwise the 'FAIL:' lines tend to scroll off screen).

Anyway, the failed tests were:

- ccnoco.test
- dirname.test (fix below)
- install2.test (same)
- lex3.test
- pr87.test

* ccnoco's second check fails (-c -o used)
* because of the way dirname.test runs the test scriptlet, it
  gets passed through command.com before getting sent to bash;
  this causes a warning from command.com to get into the output.
  The attached patch explicitly runs $SHELL, passing it the
  script.
* install2.test assumes a file is inaccessible after a 'chmod 000'
  However, on DOS/Windows, the read bit can't be cleared, so
  make dist succeeds.  I've added a test to see if chmod 000
  actually makes the file unreadable, skipping the test if it
  didn't.
* lex3.test fails because it uses a fake prefix, causing configure
  to not load its config.site, and on DJGPP the config.site is
  required for configure to work properly (otherwise, no
  executables are found, causing configure to fail because it
  can't find a C compiler)
  This is an actual problem with the 'make dist' machinery.
* in pr87.test, make distcheck seems unable to find the README
  file, even though VPATH is set to .. (where README resides)

2001-07-21  Tim Van Holder  <address@hidden>

        * tests/dirname.test: Explicitly use $SHELL to run the
        script; this avoids failures on DJGPP.
        * tests/install2.test: Skip if 'chmod 000' doesn't make
        a file unreadable.

tests-fix-1.diff
Description: Binary data