[issue21118] str.translate is absurdly slow in majority of use cases (takes up to 60x longer than similar functions)

[Serhiy Storchaka]

Serhiy Storchaka added the comment:

Previous patch and results were against source code before committing fast_translate.patch. Here is a patch synchronized with current code.

translate_timing.py results:

                                unpatched           patched
Testing 1-1 translation
str.translate                   0.5265426559999469  0.6120695240006171
str.translate from bytes trans  0.2608634099997289  0.32327288200031035
Testing deletion
str.translate                   4.331346814999051   0.7960810519998631
Testing enlarging translations
str.translate                   4.392650978999882   0.9280614529998275

translate_script_ascii.py results:

---------------------------+------------------------------+-------------------------------
Tests                      | translate_script_ascii.fastA | translate_script_ascii.cached2
---------------------------+------------------------------+-------------------------------
replace none, length=10    |                  1.54 us (*) |                 2.07 us (+34%)
replace none, length=10**3 |                  10.5 us (*) |                        10.6 us
replace none, length=10**6 |                      12.6 ms |                    12.3 ms (*)
replace 10%, length=10     |                  1.69 us (*) |                 2.31 us (+37%)
replace 10%, length=10**3  |                  10.6 us (*) |                        10.7 us
replace 10%, length=10**6  |                      12.6 ms |                    12.1 ms (*)
replace 50%, length=10     |                  1.69 us (*) |                 2.31 us (+36%)
replace 50%, length=10**3  |                  10.6 us (*) |                        10.9 us
replace 50%, length=10**6  |                      12.6 ms |                    12.3 ms (*)
replace 90%, length=10     |                  1.69 us (*) |                 2.26 us (+34%)
replace 90%, length=10**3  |                  10.6 us (*) |                        10.7 us
replace 90%, length=10**6  |                      12.6 ms |                    12.2 ms (*)
replace all, length=10     |                  1.07 us (*) |                 1.68 us (+56%)
replace all, length=10**3  |                  9.28 us (*) |                        9.46 us
replace all, length=10**6  |                      12.6 ms |                      12 ms (*)
---------------------------+------------------------------+-------------------------------
Total                      |                        63 ms |                    60.9 ms (*)
---------------------------+------------------------------+-------------------------------

translate_cached_2.patch little slows down translation of short ASCII strings (due to cache initialization and freeing), but speeds up deletion and enlarging and works with non-ASCII data.

----------
Added file: http://bugs.python.org/file34736/translate_cached_2.patch

_______________________________________
Python tracker <report at bugs.python.org>
<http://bugs.python.org/issue21118>
_______________________________________