GitLab

bugfix in check_stderr_ok()

Run the compiler from the directory in which the test source lives.
This allows simple multi-module tests using the vanilla test macros to
work (not strictly the right way to do it, but several of our tests
are of this kind).

More tidyups and code sharing.

Run each test in its own directory, rather than trying to run them all
from the directory in which the driver was started.  This is required
because several tests expect to be able to open auxilliary files in
the current directory.

expected failure fixes

don't forget to include std-macros.T

Allow expected failures.

print contents of the file that should have been empty, in verbose mode

Print a helpful message when the program produced some output and it
shouldn't have.

Allow $stdin to be overriden.

Use say_fail_if_noteq() rather than inline version.

Track some minor changes to the syntax.

Allow compiling to .hc only.

Don't try to handle multiple sample stdout/stderr files for now.

- print diffs when output files differ (in verbose mode)

- allow new output to be "accepted" by saying accept=True
  on the cmd line

- abstract some command names (to the top of the file for now)

Add a verbose flag (use ./runtests verbose=True to get the actual
command lines printed out)

Fix bogons from last commit.

-- Use diff to show diffs between actual and expected compiler output.
-- Show compiler output in cases where none was expected.

Rename some functions used in vanilla-test.T to be more consistent/logical.

Fix bug in vanilla-comp-test-actions

Improved diagnostics when a should_compile test fails.

More diagnostic wibbles.

Further improvements to printing of diagnostic information.

normalise_errmsg()s in a few more places.

Further adventures to do with approximate matching of compiler error
messages.  Use the new pipe facility in the test driver to normalise
error messages before comparison, by piping them through new program
normalise_errmsg.  This avoids various kinds of tiresome comparison
failures to do with capitalisation, whitespace changes, and path
names.

Add support for compile-only (should_run) style tests.

Track recent hackery.

Commit hackery so far.

Changes to support running "vanilla" tests -- those with a single source
file, reference .stdin, .stdout and maybe .stderr files.  These constitute
the vast majority of GHC tests.