Installing a new compiler test
==============================

[If the test if for the driver, the parser, the runtime system, the
std prelude, ... in short _not_ the compiler, it belongs elsewhere.]

1.  Copy your Haskell program into an appropriately named file in the
    appropriate directory, e.g., "typecheck/tc093.hs" for the 93rd
    typechecker test.

2.  Edit the Jmakefile in that dir to add your test: almost certainly
    just a line of the form...

RunStdTest(tc093,$(TESTGHC), "-ddump-tc tc093.hs", 0, /dev/null, tc093.stderr)
           ^^^^^
name of test||||
                 ^^^^^^^^^^
driver to use----||||||||||      	     	 
                             ^^^^^^^^^^^^^^^^^^^^
driver command line----------||||||||||||||||||||
                                                  ^^
expected exit status (0=success, 1=failure)-------||
                                                     ^^^^^^^^^
file holding expected standard output----------------|||||||||
								^^^^^^^^^^^^^
file holding expected output on standard error------------------|||||||||||||

    The example above is typical.  The command-line stuff may vary,
    but it's likely to be "dump pass <foo> output" (e.g., -ddump-tc)
    and the input file name.  Dump output is on stderr, hence the
    expected-output files.

    The current best documentation of the flags to use is in
    $(TOP)/driver/ghc (invoke w/ -help option).

3.  Create the expected-output files.  I'm usually lazy and just
    "touch" them (creating an empty file), then update them (section
    below) after the test has "failed".

4.  "make Makefile", to make a Makefile from the Jmakefile.

5.  "make runtest_<testname>" (e.g., make runtest_tc093) to run the
    one test.

    IF "make" FALLS OVER, THEN IMMEDIATEDLY "mv Makefile.bak Makefile"!!
    You probably had a typo in the Jmakefile; fix it and resume from
    step 4.


Running tests
=============

* You may run all tests by typing "make runtests" (or, if you expect
  or don't mind errors, "make -k runtests").

* You may run one test <foo> with "make runtest_<foo>".

* You may run tests <foo> to <bar> with a simple script,
  "dotests <foo> <bar>".  You may pass "make" arguments to it as well,
  as in:
    	dotests -k tc019 tc028


Updating the "expected output" files
====================================

Sometimes, it will happen that the differences between expected and
actual output of the tests will not mean failure but that the actual
output is "more correct".

If you save the output of "make runtests" (mainly from "diff"), you
may automatically update the expected-output files by using

    patch -p0 < saved-output

(You should probably ^C out of the "patch" if it doesn't do exactly
what you expect.)