Reorganize tags on benchmarks

まず issue 208 を確認し、次に既存のベンチマークタグがどこで使用されているか、またそれらを変更した場合に他のコンシューマーが影響を受けるかどうかを調査します。提案された workload、feature、web タグを含むタグの分類体系について合意し、変更による影響を文書化できれば完了です。

Once https://github.com/python/pyperformance/issues/208 is complete, we will probably want to reorganize the benchmarks into tags so that they are more useful and meaningful. This issue can hopefully provide a place for discussion.

Most importantly: Are there other places where the tags are used such that changing them would be an issue? How can I identify these folks other than posting here?

The currently assigned tags are:

apps: {'2to3', 'tornado_http', 'html5lib', 'chameleon'}
math: {'pidigits', 'float', 'nbody'}
regex: {'regex_v8', 'regex_compile', 'regex_effbot', 'regex_dna'}
serialize: {'xml_etree_generate', 'pickle_dict', 'json_dumps', 'unpickle_pure_python', 'unpickle', 'xml_etree_process', 'json_loads', 'pickle', 'xml_etree_parse', 'pickle_list', 'xml_etree_iterparse', 'unpickle_list', 'pickle_pure_python'}
startup: {'python_startup_no_site', 'python_startup'}
template: {'genshi_xml', 'mako', 'genshi_text', 'django_template'}

For the most part, I think the existing tags are fine, though apps is perhaps a little vague and perhaps should be removed.

I would propose adding the following tags (each benchmark can have multiple tags):

• Size:
  • workload: This would be for benchmarks that represent real world workloads. These would roll up into "one big number" that we report in places like the CPython release notes. I'm not crazy about the name of this tag. Suggestions?
  • feature: The opposite of a macrobenchmark, for benchmarks that test a very specific feature.
• Domain:
  • web: Typical tasks used in server-side web development: for example, serializing/deserializing HTML, JSON, XML, l10n and i18n related things