[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: using ASCII names to avoid overhead
From: |
Pádraig Brady |
Subject: |
Re: using ASCII names to avoid overhead |
Date: |
Sun, 13 Nov 2016 12:30:56 +0000 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.2.0 |
On 13/11/16 08:45, Niels Möller wrote:
> Pádraig Brady <address@hidden> writes:
>
>> I was just looking at the overhead of proper_name_utf8() vs proper_name()
>> in coreutils, and noticed that using the former adds about 18K to each
>> program.
>
> And that's code, not rodata?
Yes mostly. 14.5K extra .text, 200 extra rodata, + other stuff.
size(1) is accumulating that to 18K extra 'text'.
> >> For example truncate(1) is 62K with and 44K without.
>> I'm going to simplify usages for my name at least.
>> I don't care and anyway no one can pronounce it no matter which form is used
>> :)
>> I was wondering if you preferred to keep utf8 usage for your names?
>> These names are seen in `cat --version` output etc.
>
> I guess only factor --version is affected in my case, while Torbjörn has
> been involved in more parts of coreutils.
>
> I'd prefer to keep the dots on my name at least for display in utf8
> locales, and preferably (but less important) also in latin1 locales.
>
> I understand you want to reduce the overhead, but maybe there's some other
> way to do it? E.g,
>
> * Produce utf8 and ascii versions of the names, or the complete
> --version message, at compile time, and display one of the versions
> depending on the locale? Should be < 1K overhead. Possibly also
> support utf8 -> latin1 conversion, which is cheap provided the utf-8
> text uses precomposed characters, and likely to fit within 1K overhead
> (but maybe not worth the effort given that the latin1 world is mostly
> migrated to utf8 by now).
OK cool. I'll maybe look at adding proper_name_only_utf8()
thanks,
Pádraig