summaryrefslogtreecommitdiff
path: root/test/regress/run_regression.py
diff options
context:
space:
mode:
Diffstat (limited to 'test/regress/run_regression.py')
-rwxr-xr-xtest/regress/run_regression.py24
1 files changed, 5 insertions, 19 deletions
diff --git a/test/regress/run_regression.py b/test/regress/run_regression.py
index 1114b4a2c..6f6edf058 100755
--- a/test/regress/run_regression.py
+++ b/test/regress/run_regression.py
@@ -169,24 +169,10 @@ def run_regression(unsat_cores, proofs, dump, wrapper, cvc4_binary,
sys.exit('"{}" must be *.cvc or *.smt or *.smt2 or *.p or *.sy'.format(
benchmark_basename))
- # If there is an ".expect" file for the benchmark, read the metadata
- # from there, otherwise from the benchmark file.
- metadata_filename = benchmark_path + '.expect'
- if os.path.isfile(metadata_filename):
- comment_char = '%'
- else:
- metadata_filename = benchmark_path
-
- metadata_lines = None
- with open(metadata_filename, 'r') as metadata_file:
- metadata_lines = metadata_file.readlines()
-
- benchmark_content = None
- if metadata_filename == benchmark_path:
- benchmark_content = ''.join(metadata_lines)
- else:
- with open(benchmark_path, 'r') as benchmark_file:
- benchmark_content = benchmark_file.read()
+ benchmark_lines = None
+ with open(benchmark_path, 'r') as benchmark_file:
+ benchmark_lines = benchmark_file.readlines()
+ benchmark_content = ''.join(benchmark_lines)
# Extract the metadata for the benchmark.
scrubber = None
@@ -196,7 +182,7 @@ def run_regression(unsat_cores, proofs, dump, wrapper, cvc4_binary,
expected_exit_status = None
command_lines = []
requires = []
- for line in metadata_lines:
+ for line in benchmark_lines:
# Skip lines that do not start with a comment character.
if line[0] != comment_char:
continue
generated by cgit on debian on lair
contact matthew@masot.net with questions or feedback