Re: cast problem in code conversion

[Kenichi Handa]

In article <address@hidden>, Juri Linkov <address@hidden> writes:

> Kenichi Handa <address@hidden> writes:
>>  In article <address@hidden>, YAMAMOTO Mitsuharu <address@hidden> writes:
>>>  Calculation of the value of `ratio' in `code_convert_region' is
>>>  missing a cast to float.  That makes the result inaccurate.
>> 
>>>  In Emacs 21.2, the part in question was as follows:
>>>       float ratio = coding->produced - coding->consumed;
>>>       ratio /= coding->consumed;
>> 
>>>  But in Emacs 21.3,
>>>           ratio = (coding->produced - coding->consumed) / coding->consumed;
>> 
>>>  Reverting this change drastically improves performance when opening a
>>>  large binary file.  For example, opening the Emacs executable becomes
>>>  6 times faster for me.
>> 
>>  Thank you for finding this bug.  I've just installed a fix.

> Now visiting large files is much faster!  Is it possible that
> emacs-unicode which still takes too much time to open large files,
> suffers from a similar problem?

I think no.  I have not yet tuned code conversion of
emacs-unicode that much.  I think there's a lot of rooms to
improve performance in emacs-unicode for no-converion,
raw-text, utf-8 decoding/encoding.

---
Ken'ichi HANDA
address@hidden