diff options
Diffstat (limited to 'tools')
-rwxr-xr-x | tools/filter_openocd_log.py | 119 |
1 files changed, 119 insertions, 0 deletions
diff --git a/tools/filter_openocd_log.py b/tools/filter_openocd_log.py new file mode 100755 index 0000000..b7ae6e0 --- /dev/null +++ b/tools/filter_openocd_log.py @@ -0,0 +1,119 @@ +#!/usr/bin/env python + +import sys + +# This function is the only OpenOCD-specific part of this script. +def make_canonical(line): + # Remove the line number and time stamp. + parts = line.split(None, 3) + if len(parts) > 3: + return "%s - - %s" % (parts[0], parts[3]) + else: + return line + +def buf_startswith(buf, sequence): + if len(buf) < len(sequence): + return False + for i, entry in enumerate(sequence): + if entry[1] != buf[i][1]: + return False + return True + +def shorten_buffer(outfd, buf, current_repetition): + """Do something to the buffer to make it shorter. If we can't compress + anything, then print out the first line and remove it.""" + length_before = len(buf) + + if current_repetition: + while buf_startswith(buf, current_repetition[0]): + del buf[:len(current_repetition[0])] + current_repetition[1] += 1 + if len(buf) < length_before: + return current_repetition + outfd.write("## The following %d lines repeat %d times:\n" % ( + len(current_repetition[0]), current_repetition[1])) + for entry in current_repetition[0]: + outfd.write("# %s" % entry[1]) + + # Look for repeated sequences... + repetitions = [] + for length in range(1, len(buf)/2): + # Is there a repeating sequence of `length` lines? + matched_lines = 0 + for i, entry in enumerate(buf[length:]): + if entry[1] == buf[i % length][1]: + matched_lines += 1 + else: + break + if matched_lines >= length: + repetitions.append((matched_lines + length, length)) + + if repetitions: + repetitions.sort(key=lambda entry: (entry[0] * (entry[1] / entry[0]), -entry[1])) + matched_lines, length = repetitions[-1] + repeated = matched_lines / length + if repeated * length >= 3: + sequence = buf[:length] + del buf[:repeated * length] + + if matched_lines == length_before: + # Could be continued... + return [sequence, repeated] + + else: + outfd.write("## The following %d lines repeat %d times:\n" % + (length, repeated)) + for entry in sequence: + outfd.write("# %s" % entry[1]) + return None + + if len(buf) >= length_before: + line, _ = buf[0] + outfd.write(line) + buf.pop(0) + + if length_before <= len(buf): + print "Buffer:" + for entry in buf: + print "%r" % entry[0] + assert False + + return None + +def compress_log(infd, outfd, window): + """Compress log by finding repeated runs of lines. Runs in O(lines * + window**2), which can probably be improved.""" + # Contains line, canonical tuples + buf = [] + current_repetition = None + + for line in infd: + buf.append((line, make_canonical(line))) + if len(buf) > window: + current_repetition = shorten_buffer(outfd, buf, current_repetition) + + while len(buf) > 0: + current_repetition = shorten_buffer(outfd, buf, current_repetition) + +def main(args): + import argparse + parser = argparse.ArgumentParser( + description='Combine repeated OpenOCD debug output lines. This is ' + 'very helpful when looking at verbose log files where e.g. target ' + 'polling is repeated over and over.', + epilog='If no files are specified, read standard input.', + formatter_class=argparse.ArgumentDefaultsHelpFormatter) + parser.add_argument('file', nargs='*', help='input file') + parser.add_argument('-o', '--output', help='output file', default=sys.stdout) + parser.add_argument('-w', '--window', type=int, default=100, + help='number of lines to consider when looking for repetitions') + args = parser.parse_args(args) + + if args.file: + for f in args.file: + compress_log(open(f, "r"), args.output, args.window) + else: + compress_log(sys.stdin, args.output, args.window) + +if __name__ == '__main__': + sys.exit(main(sys.argv[1:])) |