Re: std C++ [was: [Patch] Psiconv TextEd and Word importer]


Subject: Re: std C++ [was: [Patch] Psiconv TextEd and Word importer]
From: Vlad Harchev (hvv@hippo.ru)
Date: Thu Dec 21 2000 - 03:13:08 CST


On Wed, 20 Dec 2000, Mike Nordell wrote:

 Mike, I think all this is useless. All reasonable OSes load pages of the
libraries only when they are needed from disk. So, if user doesn't use BIG5
encoding, the pages with these tables won't be loaded into RAM at all (if they
are read-only). And I have an impression that libiconv could be configured to
not support some particluar encodings at compile time. Currently non-latin1
chars are only supported on Unix, so all other platforms may omit support for
all but 2 charset (iso-8859-1 and cp1252) from libiconv (or at least omit all
CJK encodings since they take 90% of the libiconv's size).

> Thomas Fletcher wrote:
> > Great idea ... but what do we really gain?
>
> A guess would be 200-500KB binary size and *no* RAM (virtual memory) usage
> unless these specific features are used (like BIG5 if I'm not completely
> mistaken) that would gain 99% percent of our user I think.
>
> [...]
> > Gains ... losses ... tough to measure
>
> True, without any real-world evidence I can't make a point. Perhaps I should
> create some empirical data? OK, I'll make a test-implementation of this, but
> it won't be until 2001, I'm off a week from tomorrow... ;-)
>
> I was thinking of a really simple thingie: only touch the (read-only) data
> once we need it, and then decompress it in one sweep. RAM usage would *then*
> become what it is now, but it would never be what it is today for most users
> (since ost users don't use these charsets). I might be wrong though, perhaps
> it's better to just run the executable through e.g. UPX (but then no
> embedded platform would ever benefit from my proposed stuff)..
>
> > but I think that there are still some other things that we
> > should probably look at before going this route.
>
> Oh yes, we have many pieces to look at. I think there are many (not just
> "some") things we have to fix before we even think 0.7.14. We have (from my
> POV) equally many things to check into before 0.7.13.
[...]

 Best regards,
  -Vlad



This archive was generated by hypermail 2b25 : Thu Dec 21 2000 - 04:48:42 CST