Re: [xml] setting the default charset ?

[Cyrille Chepelov <chepelov calixo net>]

Le lun, jui 30, 2001, à 09:08:46 -0400, Daniel Veillard a écrit:

>   This was well known, impossible to fix in libxml1 and fixed for nearly
> 1.5 years in libxml2. The NO EXCUSE rule applies ... I know it applies to the 
> whole Gnome project. 

And I've just recently learned that it will be tougher than I previously 
anticipated to convert dia to libxml2, because some builds require bonobo 
and gnome-print stuff, and these will link in libxml1. 

>   To come back to Dia, guessing is the last thing to try. Ask the user if
> you have a trouble, BUT DO NOT GUESS ! If you guess and load a russian encoded
> file in a french environment based on the local and that the file is saved
> then you get a successful but wrong conversion and if the user saves the
> document all non-ascii will be wrong permanently.

We will guess. We will warn, and perhaps take some protective measures so
that the innocent doesn't get harmed (like, saving the original broken file 
in a .dia.brokenxml file). 

Presenting the result of `iconv -l` in a list in a dialog box, asking the
user to choose the right one is as productive as writing French using
Ugaritic or Maya script and expecting the user to understand what's going to
happen. You know what an encoding system is ; 99.99% of users don't [except
perhaps Cyrillic script users which have to deal with different
charsets on the web on just every script, and some Chinese users to perhaps
a lesser extent]

(It is no use even writing the GUI to let an "advanced" user choose the
right encoding system ; "advanced" users are advanced enough to follow a FAQ
telling them how to gunzip and edit an XML file if the chosen default is
wrong. And I'm all in favour of a pure Darwinian solution (there is 230VAC
available in a typical PC, isn't it ?) for users who blindly click on
"OK"-style dialog boxes ).

(I've implemented the awkward scheme I wrote about yesterday ; the resulting
code works whether saturday's patch is included or not. Which is better,
since versioned dependencies cause mandatory upgrades, mandatory upgrades
cause fear, etc. Well, it's still a useless extra gzopen().).

        -- Cyrille

-- 
Grumpf.