gen_bench_expectations to generate bench expectations files.

bench/bench_util.py

 '''
 Created on May 19, 2011
 
 @author: bungeman
 '''
 
+import os
 import re
 import math
 
 # bench representation algorithm constant names
 ALGORITHM_AVERAGE = 'avg'
 ALGORITHM_MEDIAN = 'med'
 ALGORITHM_MINIMUM = 'min'
 ALGORITHM_25TH_PERCENTILE = '25th'
 
 # Regular expressions used throughout.
@@ skipping 123 matching lines @@
         for new_time in TIME_RE_COMPILED.finditer(times):
             current_time_type = new_time.group(1)
             iters = [float(i) for i in
                      new_time.group(2).strip().split(',')]
             value_dic.setdefault(bench, {}).setdefault(
                 current_config, {}).setdefault(current_time_type, []).append(
                     iters)
             layout_dic.setdefault(bench, {}).setdefault(
                 current_config, {}).setdefault(current_time_type, tile_layout)
 
+def parse_skp_bench_data(directory, revision, rep, default_settings=None):
+    """Parses all the skp bench data in the given directory.
+
+    Args:
+      directory: string of path to input data directory.
+      revision: git hash revision that matches the data to process.
+      rep: bench representation algorithm, see bench_util.py.
+      default_settings: dictionary of other run settings.

[borenet, 2014/03/17 19:14:42]

Can you give some examples of what these run settings might be?

[benchen, 2014/03/17 19:37:54]

Added some. Seems like this is currently only used by microbenches, but we may
activate for picture benches, or add microbench in parsing later. Shall I remove
it for now?
On 2014/03/17 19:14:42, borenet wrote:

[borenet, 2014/03/17 20:07:26]

hmm.. Are all of the acceptable settings documented somewhere?  Maybe the best
thing to do is to point to that, if it exists.  I'll leave it up to you whether
to remove it or point to more documentation.

+
+    Returns:
+      A list of BenchDataPoint objects.
+    """
+    if not default_settings:
+      default_settings = {}
+
+    revision_data_points = []
+    file_list = os.listdir(directory)
+    file_list.sort()
+    for bench_file in file_list:
+        scalar_type = None
+        # Scalar type, if any, is in the bench filename after 'scalar_'.
+        if (bench_file.startswith('bench_' + revision + '_data_')):
+            if bench_file.find('scalar_') > 0:
+                components = bench_file.split('_')
+                scalar_type = components[components.index('scalar') + 1]
+        else:  # Skips non skp bench files.
+            continue
+
+        with open('/'.join([directory, bench_file]), 'r') as file_handle:
+          settings = default_settings

[borenet, 2014/03/17 19:14:42]

This will point to the same object and will therefore still modify the input.  I
suggest something like:
settings = dict(default_settings or {})

And then you can remove the "if not default_settings" above.

[benchen, 2014/03/17 19:37:54]

ah right. fixed.
On 2014/03/17 19:14:42, borenet wrote:

+          settings['scalar'] = scalar_type
+          revision_data_points.extend(parse(settings, file_handle, rep))
+
+    return revision_data_points
+
 # TODO(bensong): switch to reading JSON output when available. This way we don't
 # need the RE complexities.
 def parse(settings, lines, representation=None):
     """Parses bench output into a useful data structure.
 
     ({str:str}, __iter__ -> str) -> [BenchDataPoint]
     representation is one of the ALGORITHM_XXX types."""
 
     benches = []
     current_bench = None
@@ skipping 153 matching lines @@
     """
     return '<a href="http://code.google.com/p/skia/source/detail?r=%s">%s</a>'%(
         revision_number, revision_number)
 
 def main():
     foo = [[0.0, 0.0], [0.0, 1.0], [0.0, 2.0], [0.0, 3.0]]
     LinearRegression(foo)
 
 if __name__ == "__main__":
     main()