abinit/tests/pymods/fldiff.py

595 lines
22 KiB
Python
Executable File

"""
Compare 2 output files from ABINIT line by line with arithmetic
comparison of floating point substrings
The first character of each line in both files indicates the mode of
comparison; these first characters MUST be identical in corresponding lines.
By default, a floating point comparison with a tolerance of 1.01d-10 is done on
all floating point strings and a character comparison is done on all other
strings, disregarding multiple spaces. For compatibility reasons with
historical fldiff.pl a float is must contains a dot and at least one digit
after it to be recognized as float. In order for two numbers to be declared
different, BOTH the absolute difference and the relative difference (difference
divided by the sum of absolute values) must be bigger than the tolerance.
Some special characters at the beginning of lines require a different handling:
- mark lines as same regardless to their content (i.e. ignore lines)
(can be be present in the 2 files or not, but must begin with -)
_ mark lines as same regardless to their content
(must be present in the 2 files, but can begin with _ in only one of them)
+ mark lines as different regardless to their content
, handle as + if ignore option is False and as - else
P handle as + if ignoreP option is False and as - else
% floating point comparisons are done with a tolerance of 1.01e-2
; floating point comparisons are done irrespective of signs
: ignore floating point numbers and do a characters comparison
. do a characters comparison, but do not count this line in the Summary
) floating point comparison is done only for the FIRST numeric field
Both files should have the same number of non - starting lines.
The ignore options affects the treatment of the ',' special character in the
first column (see above)
The ignoreP options affects the treatment of the 'P' special character in the
first column (see above)
The label option, if specified, is appended at the end of the summary.
The tolerance option set the tolerance for comparison of floats, the default
is 1.01e-10. This modifications do not apply to the tolerance determined by
the '%',and '.' first-column special signs.
"""
from __future__ import print_function, division, unicode_literals
import re
from math import floor
from threading import Thread
from .data_extractor import DataExtractor
from .yaml_tools import is_available as has_yaml
if has_yaml:
from .yaml_tools.driver_test_conf import DriverTestConf as YDriverConf
from .yaml_tools.tester import Tester as YTester, Failure as YFailure
# Match floats. Minimal float is .0 for historical reasons.
# As a consequence, integers will be compared as strings
float_re = re.compile(r'([+-]?[0-9]*\.[0-9]+(?:[eEdDfF][+-]?[0-9]+)?)')
def norm_spaces(s):
r"""Normalize all blanks ( \\n\\r\\t)."""
# the join/split technique remove all blanks and put one space between non-blanks words
return ' '.join(s.split())
def relative_truncate(f, n):
"""
>>> rel_truncate(1.8367387367, 2)
1.83
>>> rel_truncate(1.8367387367e-5, 5)
1.83673e-05
>>> rel_truncate(1.8367387367e+7, 4)
18367000.0
"""
ten_n = 10.0**n
if f == 0.0:
return 0.0
elif abs(f) >= 1:
ten_p = 10.0
while abs(f) > ten_p:
ten_p *= 10.0
fact = 10 * ten_n / ten_p
return floor(f * fact) / fact
else:
ten_p = 0.1
p = -1
while abs(f) < ten_p:
ten_p *= 0.1
p -= 1
fact = ten_n / ten_p
return floor(f * fact) / fact
class NotDriverConf(object):
def __init__(self, has_yaml):
self.has_yaml = has_yaml
def extra_info(self):
if self.has_yaml:
return ('# YAML support is available, but is disabled for this test.',)
else:
return ('# YAML support is not available, YAML based tests will be ignored.',)
class LineDifference(object):
"""Base class representing a difference."""
def __init__(self, p1, p2, l1, l2):
self.lines = (p1 + 1, p2 + 1)
if l1 == '' or l1[-1] not in '\n\r':
l1 += '\n'
self.content = (l1, l2)
def __eq__(self, other):
"""Implement the == test."""
return self.lines == other.lines and self.content == other.content
def __ne__(self, other):
return not (self == other)
def __repr__(self):
"""Default representation of difference inspired by gnu diff tool."""
return (
'{}\n'.format(*self.lines)
+ '< ' + self.content[0]
+ '> ' + self.content[1]
)
class LineCountDifference(LineDifference):
"""Represents a difference between line counts."""
def __init__(self, more, less, line_count=(0, 0)):
"""
Args:
more: the name of the file with more lines
less: the name of the file with less lines
"""
LineDifference.__init__(self, 0, 0, '', '')
self.more = more
self.less = less
self.line_count = line_count
def __repr__(self):
if self.line_count != (0, 0):
return ('{} has more significant lines than {} ({} > {}).\n'
.format(self.more, self.less, *self.line_count))
return '{} has more significant lines than {}.\n'.format(self.more, self.less)
class MetaCharDifference(LineDifference):
"""Represent a difference between two lines with different meta characters."""
def __init__(self, p1, p2, m1, m2):
LineDifference.__init__(self, p1, p2, '', '')
self.metas = (m1, m2)
def __repr__(self):
return ('At line {} (in file 1), line {} (in file 2), different'
' leading characters: `{}` and `{}`.\n').format(*(self.lines + self.metas))
class FloatDifference(LineDifference):
"""Represent a difference between floating point values."""
def __init__(self, p1, p2, line1, line2, abs_err, rel_err):
LineDifference.__init__(self, p1, p2, line1, line2)
self.abs_err = abs_err
self.rel_err = rel_err
class TextDifference(LineDifference):
"""Represent a difference between text parts of a lines."""
def __init__(self, p1, p2, line1, line2, silent=False):
LineDifference.__init__(self, p1, p2, line1, line2)
self.silent = silent
class ForcedDifference(LineDifference):
"""A difference that is arbitrarly declared."""
class Result(object):
"""Analyse and summarize the set of differences found by a diff."""
def __init__(self, fl_diff, yaml_diff, extra_info=[], label=None, verbose=False):
"""
differences is expected to be a list of Difference instances
"""
self.fl_diff = fl_diff
self.yaml_diff = yaml_diff
self.extra_info = extra_info
self.fatal_error = False
self.yaml_error = False
self.success = True
self.max_abs_err = 0.0
self.max_rel_err = 0.0
self.max_abs_ln = 0
self.max_rel_ln = 0
self.ndiff_lines = 0
self.label = label
self.verbose = verbose
self.details = self._analyse()
def _analyse(self):
"""
Analyse a difference list and extract summary information and
details. Summary information is
- self.max_abs_err: maximum absolute difference
- self.max_rel_err: maximum relative difference
- self.max_abs_ln: line number where the maximum absolute
difference is reached for the first time
- self.max_rel_ln: line number where the maximum relative
difference is reached for the first time
- self.ndiff_lines: number of lines flagged as different (excluding
"silent" differences: line starting with '.' '+' and depending of Diff options ',' and 'P')
"""
details = []
error_lines = set()
if self.yaml_diff:
details.append('# Start YAML based comparison report\n')
for diff in self.yaml_diff:
if diff.is_fail():
self.success = False
self.yaml_error = True
details.append(repr(diff) + '\n\n')
elif self.verbose:
details.append(repr(diff) + '\n\n')
if self.fl_diff:
details.append('# Start legacy fldiff comparison report\n')
for diff in self.fl_diff:
if isinstance(diff, LineCountDifference) \
or isinstance(diff, MetaCharDifference):
self.fatal_error = True
self.success = False
details = str(diff)
elif isinstance(diff, ForcedDifference):
pass # Silent differences: not counted as different line
elif isinstance(diff, FloatDifference):
if diff.lines[0] not in error_lines:
self.ndiff_lines += 1
if diff.abs_err > self.max_abs_err:
self.max_abs_err = diff.abs_err
self.max_abs_ln = diff.lines[0]
if diff.rel_err > self.max_rel_err:
self.max_rel_err = diff.rel_err
self.max_rel_ln = diff.lines[0]
self.success = False
elif isinstance(diff, TextDifference):
if not diff.silent:
if diff.lines[0] not in error_lines:
self.ndiff_lines += 1
self.success = False
else: # any other Difference
assert isinstance(diff, LineDifference), 'Unknown type of Difference.'
if diff.lines[0] not in error_lines:
self.ndiff_lines += 1
self.success = False
if self.fatal_error:
break # stop analysing in case of fatal error
if diff.lines[0] not in error_lines:
details.append(str(diff))
error_lines.add(diff.lines[0])
return details
def get_summary(self):
"""Return a textual summary of the diff."""
if self.yaml_error:
summary = 'yaml_test errors.'
elif self.fatal_error:
summary = 'fldiff fatal error.'
elif self.success:
summary = 'no significant difference has been found.'
else:
summary = ('different lines={}, max abs_diff={:.3e} (l.{}),'
' max rel_diff={:.3e} (l.{}).').format(
self.ndiff_lines,
self.max_abs_err,
self.max_abs_ln,
self.max_rel_err,
self.max_rel_ln
)
if self.label is not None:
summary = 'Summary ' + self.label + ': ' + summary
else:
summary = 'Summary: ' + summary
return summary
def dump_details(self, file=None):
"""
Either return a string describing all detected differences
or write it into the given file (expected to be a writable stream).
"""
if file is None:
return ('\n'.join(self.extra_info) + '\n' + ''.join(self.details)
+ self.get_summary())
else:
file.write('\n'.join(self.extra_info) + '\n')
file.writelines(self.details)
file.write(self.get_summary() + '\n')
return None
def passed_within_tols(self, tolnlines, tolabs, tolrel):
"""
Check the result of the diff against the given tolerances.
"""
if self.yaml_error:
status = 'failed'
for diff in self.yaml_diff:
if diff.is_fail():
first_fail = diff
break
msg = 'yaml_test errors. First is:\n{}\n'.format(first_fail)
elif self.fatal_error:
status = 'failed'
msg = 'fldiff fatal error:\n' + self.details
elif self.success:
status = 'succeeded'
msg = 'succeeded'
else:
# truncate to prevent fldiff from printing 1.000 < 1.000
# compatibility fix, this may be removed later
abs_error = relative_truncate(self.max_abs_err, 3)
rel_error = relative_truncate(self.max_rel_err, 3)
ndiff_lines = self.ndiff_lines
status = 'failed'
fact = 1.0
locs = locals()
if ndiff_lines > tolnlines:
msg = 'failed: erroneous lines {ndiff_lines} > {tolnlines}'
elif abs_error > tolabs * fact and rel_error < tolrel:
msg = 'failed: abs error {abs_error:.4} > {tolabs}'
elif rel_error > tolrel * fact and abs_error < tolabs:
msg = 'failed: rel error {rel_error:.4} > {tolrel}'
elif abs_error > tolabs * fact and rel_error > tolrel * fact:
msg = ('failed: abs error {abs_error:.4} > {tolabs},'
' rel error {rel_error:.4} > {tolrel}')
else:
status = 'passed'
msg = ('passed: abs error {abs_error:.4} < {tolabs},'
' rel error {rel_error:.4} < {tolrel}')
msg = msg.format(**locs)
isok = status in ('succeeded', 'passed')
return isok, status, msg
def has_line_count_error(self):
return any(isinstance(diff, LineCountDifference)
for diff in self.fl_diff)
class Differ(object):
def __init__(self, yaml_test=None, **options):
"""
Init a differ with some parameters passed via options.
Known parameters are:
- ignore: bool (default True)
- ignoreP: bool (default True)
- tolerance: float (tolerance for both relative and absolute
difference)
- tolerance_abs: float (default 1.01e-10)
- tolerance_rel: float (default 1.01e-10)
- label: str (default None)
- use_yaml: bool (default True)
- use_fl: bool (default True)
- verbose: bool (default False) enable report of successful yaml tests too
"""
self.xml_mode = False # this is the first dirty fix.
self.options = {
'ignore': True,
'ignoreP': True,
'tolerance_abs': 1.01e-10,
'tolerance_rel': 1.01e-10,
'label': None,
'use_fl': True,
'use_yaml': False,
'verbose': False,
'debug': False
}
self.options.update(options)
if 'tolerance' in options:
self.options['tolerance_abs'] = options['tolerance']
self.options['tolerance_rel'] = options['tolerance']
self.use_fl = self.options['use_fl']
self.use_yaml = has_yaml and self.options['use_yaml']
if self.use_yaml:
if yaml_test and 'file' in yaml_test and yaml_test['file']:
self.yaml_conf = YDriverConf.from_file(yaml_test['file'])
elif yaml_test and 'yaml' in yaml_test and yaml_test['yaml']:
self.yaml_conf = YDriverConf(yaml_test['yaml'])
else:
self.yaml_conf = YDriverConf()
self.yaml_conf.debug = self.options['debug']
else:
self.yaml_conf = NotDriverConf(has_yaml)
def diff(self, file1, file2):
"""
Compute the diff of file 1 (reference) and file 2 (out)
and return a Result instance.
"""
if file1.endswith('.xml'):
self.xml_mode = True
with open(file1, 'rt') as f1, open(file2, 'rt') as f2:
line_diff, doc_diff = self._diff_lines(f1, f2)
return Result(line_diff, doc_diff,
extra_info=self.yaml_conf.extra_info(),
label=self.options['label'],
verbose=self.options['verbose'])
def _diff_lines(self, src1, src2):
lines = [None, None]
documents = [None, None]
corrupted = [None, None]
def extractor(src, i):
# Remark: self.options['use_yam'] -> explicit request for YAML
# self.use_yaml -> explicit request AND availability of YAML
dext = DataExtractor(self.options['use_yaml'], xml_mode=self.xml_mode,
ignore=self.options['ignore'], ignoreP=self.options['ignoreP'])
lines[i], documents[i], _ = dext.extract(src)
corrupted[i] = dext.corrupted_docs
t1 = Thread(target=extractor, args=(src1, 0))
t2 = Thread(target=extractor, args=(src2, 1))
t1.start()
t2.start()
t1.join()
t2.join()
if self.use_fl:
lines_differences = self._fldiff(*lines)
else:
lines_differences = []
if not self.use_yaml:
doc_differences = []
elif corrupted[0]:
doc_differences = [YFailure(
self.yaml_conf,
'Reference has corrupted YAML documents at line(s) {}.'
.format(', '.join(str(d.start + 1) for d in corrupted[0]))
)]
elif corrupted[1]:
doc_differences = [YFailure(
self.yaml_conf,
'Tested file has corrupted YAML documents at line(s) {}.'
.format(', '.join(str(d.start + 1) for d in corrupted[1]))
)]
else:
doc_differences = self._test_doc(*documents)
return lines_differences, doc_differences
def _test_doc(self, docs1, docs2):
"""Compare docs2 to docs1 and apply tests on docs2."""
return YTester(docs1, docs2, self.yaml_conf).run()
def _fldiff(self, lines1, lines2):
"""
Compute the effective comparison between two set of lines.
LineCountDifference and MetaCharDifference are both fatal so
they are returned alone if encountered.
"""
differences = []
if len(lines1) > len(lines2):
return [LineCountDifference('file 1', 'file 2', (len(lines1), len(lines2)))]
elif len(lines1) < len(lines2):
return [LineCountDifference('file 2', 'file 1', (len(lines2), len(lines1)))]
else:
for (i1, meta1, line1), (i2, meta2, line2) in zip(lines1, lines2):
if meta1 != meta2:
if meta1 != '_' and meta2 != '_':
return [MetaCharDifference(i1, i2, meta1, meta2)]
else:
if meta1 == '_':
# ignore these lines
pass
elif meta1 == '+':
# these lines are arbitrarily different
differences.append(ForcedDifference(
i1, i2, line1, line2
))
elif meta1 in {':', '.'}:
# do a character comparison
if norm_spaces(line1) != norm_spaces(line2):
differences.append(TextDifference(
i1, i2, line1, line2, silent=(meta1 == '.')
))
else:
# compare numerical values
splitted1 = float_re.split(line1)
splitted2 = float_re.split(line2)
# not the same number of floats on the line
if len(splitted1) != len(splitted2):
differences.append(TextDifference(i1, i2, line1, line2))
else:
if meta1 == '%':
# force tolerance
tol = 1.01e-2
tolrel = tol
else:
tol = self.options['tolerance_abs']
tolrel = self.options['tolerance_rel']
def to_float(f):
return float(f.lower().replace('d', 'e').replace('f', 'e'))
def pairs(seq1, seq2):
i = 0
n = len(seq1)
while i + 1 < n:
yield (seq1[i], seq1[i + 1],
seq2[i], seq2[i + 1])
i += 2
if i < n:
yield (seq1[i], None, seq2[i], None)
# si -> plain text separators
# fi -> floats
flag_compare = 1
for s1, f1, s2, f2 in pairs(splitted1, splitted2):
if norm_spaces(s1) != norm_spaces(s2):
differences.append(TextDifference(
i1, i2, line1, line2
))
if f1 is not None:
# reached the end
f1 = to_float(f1)
f2 = to_float(f2)
if meta1 == ';':
# compare absolute values
f1, f2 = abs(f1), abs(f2)
abs_sum = abs(f1) + abs(f2)
diff = abs(f1 - f2)
if abs_sum == 0.0:
diffrel = 0.0
else:
diffrel = diff / abs_sum
if diff > tol and diffrel > tolrel and flag_compare==1:
differences.append(
FloatDifference(
i1, i2, line1, line2,
diff, diffrel
)
)
if meta1 == ')':
# turn off further numerical comparison
flag_compare = 0
return differences