Provided by: creduce_2.10.0+20200116-1ubuntu1_amd64
creduce - C and C++ program reducer
creduce 2.11.0 (unknown) -- a C and C++ program reducer C-Reduce requires an "interestingness test" and one or more files to reduce, which must be writable. The interestingness test is an executable program (usually a shell script) that returns 0 when a partially reduced file is interesting (a candidate for further reduction) and returns non-zero when a partially reduced file is not interesting (not a candidate for further reduction -- all uninteresting files are discarded). C-Reduce runs the interestingness test in a fresh temporary directory containing only the partially reduced file(s). Thus, when the interestingness test examines a partially reduced file, it must do so using a relative path to the current working directory. On the other hand, when the interestingness test refers to any file that is not being reduced, this should be done using an absolute path. The interestingness test should be deterministic and should not expect any command line arguments. C-Reduce can and does introduce infinite loops during reduction. Therefore, if your interestingness test runs the compiled program, this should be done under a timeout. You can implement the timeout yourself, for example using the UNIX "timeout" command, or alternatively you can give C-Reduce a timeout (which is on by default, but is set very long: 5 minutes). An advantage of using C-Reduce's timeout is that when a test is killed due to timing out, a message will be printed. If you do implement a timeout yourself, we recommend that your interestingness test returns exit code 124 when a subcommand times out. This will not affect the reduction but it will cause C-Reduce to print a message stating the interestingness test reported a timeout. It is important to ensure that timeouts are not occurring too often (especially during C-Reduce's initial passes) or else C-Reduce will work poorly. There is a particularly subtle issue regarding timeouts, which is that the interestingness test may slow down due to resource contention when C-Reduce runs multiple copies of it on different cores. We have observed slowdowns of up to 50% due to (we suspect) cache thrashing. You must take this slowdown into account when choosing a timeout or else timeouts might occur very frequently. As a quick example of an interestingness test, if you consider a file to be interesting if GCC's vectorizer fires while compiling it, you might write this shell script: gcc -w -O3 foo.c -S && grep xmm foo.s To see if your interestingness test is working, try running these commands: DIR=`mktemp -d` cp file_to_reduce [optionally, more files to reduce] $DIR cd $DIR /path/to/interestingness_test echo $? This should result in "0" being echoed to the terminal. If this does not happen, the interestingness test is flawed and C-Reduce won't be able to make use of it. If you haven't written an interestingness test before, please refer to this tutorial for additional guidance: https://embed.cs.utah.edu/creduce/using/ If at all possible, run C-Reduce on preprocessed code, generated for example using: gcc -E -P file.c If you cannot reduce preprocessed code, you can either reduce just the non-preprocessed file or else perform a multi-file reduction on the file and its transitive includes (or any subset of them). In the first case you need to set the CREDUCE_INCLUDE_PATH environment variable to a colonseparated list of include directories in order for clang_delta to find them. If your interestingness test involves a cross compiler and the characteristics of the cross target differs from the host you will need to set CREDUCE_TARGET_TRIPLE to match the cross target. This is particularly important if you are working with non-preprocessed code and use CREDUCE_INCLUDE_PATH. Press "s" at any time to skip to the next pass (this feature is disabled unless the Perl module Term::ReadKey is available on your system). Summary of options: --abs-timing Print timestamps about reduction progress using absolute time --add-pass <pass> <sub-pass> <priority> Add the specified pass to the schedule --also-interesting <exitcode> A process exit code (somewhere in the range 64-113 would be usual) that, when returned by the interestingness test, will cause C-Reduce to save a copy of the variant [default: -1] --debug Print debug information --die-on-pass-bug Terminate C-Reduce if a pass encounters an otherwise non-fatal problem --max-improvement <bytes> Largest improvement in file size from a single transformation that C-Reduce should accept (useful only to slow C-Reduce down) --n <N> Number of cores to use; C-Reduce tries to automatically pick a good setting but its choice may be too low or high for your situation --no-cache Don't cache behavior of passes --no-default-passes Start with an empty pass schedule --no-give-up Don't give up on a pass that hasn't made progress for 50000 iterations --no-kill Wait for parallel instances to terminate on their own instead of killing them (only useful for debugging) --not-c Don't run passes that are specific to C and C++, use this mode for reducing other languages --print-diff Show changes made by transformations, for debugging --save-temps Don't delete /tmp/creduce-xxxxxx directories on termination --shaddap Suppress output about non-fatal internal errors --skip-initial-passes Skip initial passes (useful if input is already partially reduced) --skip-key-off Disable skipping the rest of the current pass when "s" is pressed --sllooww Try harder to reduce, but perhaps take a long time to do so --tidy Do not make a backup copy of each file to reduce as file.orig --timeout Interestingness test timeout in seconds [default: 300] --timing Print timestamps about reduction progress --version Print the version information usage: creduce [options] interestingness_test file_to_reduce [optionally, more files to reduce] creduce --help for more information
The full documentation for creduce is maintained as a Texinfo manual. If the info and creduce programs are properly installed at your site, the command info creduce should give you access to the complete manual.