[Python-Dev] cpython: Add fast-path in PyUnicode_DecodeCharmap() for pure 8 bit encodings:

[Victor Stinner]

2013/4/11 Serhiy Storchaka <storchaka at gmail.com>:
> On 09.04.13 23:29, victor.stinner wrote:
>>
>> http://hg.python.org/cpython/rev/53879d380313
>> changeset:   83216:53879d380313
>> parent:      83214:b7f2d28260b4
>> user:        Victor Stinner <victor.stinner at gmail.com>
>> date:        Tue Apr 09 21:53:09 2013 +0200
>> summary:
>>    Add fast-path in PyUnicode_DecodeCharmap() for pure 8 bit encodings:
>> cp037, cp500 and iso8859_1 codecs
>
> I deliberately specialized only most typical case in order to reduce
> maintaining cost. Further optimization of two not the most popular encodings
> probably not worth additional 25 lines of code.

I did the commit while I was trying to avoid usage of
PyUnicode_READ_CHAR() and PyUnicode_READ() in unicodeobject.c (slow
macros). I was surprised that PyUnicode_DecodeCharmap() has a fast
path for Py_UCS2 mapping but not Py_UCS1 mapping.

After implementing the fast-path, I realized that only a very few codecs use it.

So what do you suggest? Revert the commit to restore the following
"hack" (to only have one fast-path)?

'\ufffe'   ## Widen to UCS2 for optimization

The Py_UCS1 fast-path has a small advantage: decoding cannot fail (no
need to call an error handler).

Victor