This file describes the flex test suite.

* WHO SHOULD USE THE TEST SUITE?

The test suite is intended to be used by flex developers, i.e., anyone hacking
the flex distribution. If you are simply installing flex, then you can ignore
this directory and its contents.

* STRUCTURE OF THE TEST SUITE

Each test is centered around a scanner known to work with the most
recent version of flex. In general, after you modify your copy of the
flex distribution, you should re-run the test suite. Some of the tests
may require certain tools to be available (e.g., bison, diff). If any
test returns an error or generates an error message, then your
modifications *may* have broken a feature of flex. At a minimum,
you'll want to investigate the failure and determine if it's truly
significant.

* HOW TO RUN THE TEST SUITE

To build and execute all tests from the top level of the flex source tree:

  $ make check

To build and execute a single test:

  $ cd tests/ # from the top level of the flex tree.
  $ make testname.log

  where "testname" is the name of the test. This is an automake-ism
  that will create (or re-create, if need be), a log of the particular
  test run that you're working on.

* HOW TO ADD A NEW TEST TO THE TEST SUITE

** If possible, just write a *.rules file.  Exit with status 1 to
   indicate failure, e.g. with a "." production if nothing
   prior properly consumed every input token.  The test machinery
   will expand a rules file into tests for all back ends.  Note that
   text following ### at the end of the ruleset will be used as input.

** Otherwise, list your test in the TESTS variable in Makefile.am in
   this directory. Note that due to the large number of tests, we use
   variables to group similar tests together. This also helps with
   handling the automake test suite requirements. Hopefully your test
   can be listed in SIMPLE_TESTS. You'll need to add the appropriate
   automake _SOURCES variable as well, and .gitignore lines for the
   binary and generated code. If you're unsure, then consult
   the automake manual, paying attention to the parallel test harness
   section.

** On success, your test should return zero.

** On error, your test should return 1 (one) and print a message to
   stderr, which will have been redirected to the log file created by the
   automake test suite harness.

** If your test is skipped (e.g., because bison was not found), then
   the test should return 77 (seventy-seven). This is the exit status that
   would be recognized by automake's "test-driver" as _skipped_.

** Once your work is done, submit a patch via the flex development
   mailing list, the github pull request mechanism or some other
   suitable means.

* NAMING CONVENTIONS

A test with an _nr suffix exercises a non-reentrant scanner built
with the default cpp back end.

A test with an _r suffix exercises a reentrant scanner built
with the default cpp back end.

A test with a c99 suffix exercises the c99 back end. All C99
scanners are re-entrant.

Most tests occur in groups with a common stem in the names, like
alloc_extra_ or debug_.  These are exercising the same token grammar
under different back ends.  As new target languages are added these
groups of parallel tests will grow. Tests that are not part of one of
these series are usually of features supported on the default cpp
back end only.

The "generated" tests are made by wrapping boilerplate code around a
rules file. The assumption is that the test binary should consume the
data following ### in the rules file, exercising some set of Flex
features and not returning a nonzero status.

If you can express your test as a ruleset within in this protocol,
please do so.  Those tests are easy to port to new back ends, as that
whole job can be done by enhancing testmaker.m4.

* WHY SOME TESTS ARE MISSING

The "top" test is backend-independent; what it's really testing
is Flex's ability to accumulate and ship preamble code sections.

The "quotes" test is also backend-independent; it's testing that
m4 expansion doesn't leak through the generated scanner.

The C99 backend is missing tests for the Bison bridge, header
generation, and loadable tables because it omits those features in
order to be a simpler starting point for wring new back ends.