blob: a24d79ed748d44d0706b2296b83d51fd6363268d [file] [log] [blame]
# This is sort of a prototype test case, which parses the listing output
# from the assembler. Later, more prototypes should be added for cases
# where objdump gets run over the .o file, and anything else like that...
# When you write a test case that uses the listing output, just copy this
# file (trimming down the overly-verbose comments a little), and
# adjust it to do what you need.
# Remember that any ".exp" file found in the tree will be processed by
# dejagnu.
proc do_foo {} {
# This string is used below when printing out a success or failure message.
# If more than one test is run by a given .exp file, it'd be nice to include
# the name of the input file.
set testname "foo.s: multi-register tweaking and frobnication"
# I use this as a flag to record whether the test case passed. If this
# flag is still clear when EOF is reached, this test fails. If there are
# two or more patterns, and I need to see all of them, I'll create N variables
# and check if the sum is N.
set x 0
# Call gas_start with two arguments: The input file name (which it'll search
# for in $srcdir/$subdir, that is, the source directory where the .exp file
# is), and a (possibly empty) string of options to pass to the assembler.
gas_start "foo.s" "-al"
# Now I just iterate over all the output lines, looking for what I want
# to see. Since each pattern explicitly will not span line breaks, there's
# also a pattern for lines that don't match anything else. (Is it safe to
# use ".*" for patterns not crossing line breaks? I don't think "$" does the
# right thing for that, in any case. I should check into whether the extra
# pattern is even needed.
# Apparently CRLF is received when using ptys for subprocesses; hence the
# \r\n for matching line number 3.
# Note that if you use "{ ... }" for the expect clause, you can't have
# comments inside it.
# This test case is kinda bogus in that seeing either a word of all zeros
# at address zero or a C-style comment on line three that says "Looking for
# C comments" (with very specific punctuation and whitespace) will cause
# it to pass this test. Usually
while 1 {
expect {
-re "^ +\[0-9\]+ 0000 00000000\[^\n\]*\n" { set x 1 }
-re "^ +3\[ \t\]+/. Looking for C comments. ./\r\n" { set x 1 }
-re "\[^\n\]*\n" { }
timeout { perror "timeout\n"; break }
eof { break }
# This was intended to do any cleanup necessary. It kinda looks like it isn't
# needed, but just in case, please keep it in for now.
# Did we find what we were looking for? If not, flunk it.
if $x then { pass $testname } else { fail $testname }
# Now actually run the test. It can be conditionalized if the test is
# not appropriate for all targets. The proc "istarget" checks a generalized
# form of the target name, so that (e.g.) "m68332-unknown-aout" would match
# here. So far, I think only the CPU name is actually ever altered.
if [istarget m68k-*] then {
# This is a tiny bit like the C compiler torture tests, in that it'll run
# the assembler with the power set of the list of options supplied.
# The first argument is the test file name; the second is arguments that
# are always to be provided; the third is a space-separated list of options
# which are optional (ending in ">" if output should be ignored, like "-a>");
# the fourth is the name of the test. So far, only binary options are handled
# this way; N-way options (like CPU type for m68k) aren't handled yet.
# The variable $stdoptlist usually has a reasonable set of optional options
# for this target.
# No, PIC isn't supported yet. This is only an example.
gas_test "quux.s" "-K" $stdoptlist "use of quuxes in PIC mode"