Re: Errors in MathML DTD

On 19 Apr 2003, William F Hammond wrote:

> >      Fetching the MathML DTD + modules by HTTP is too big an overhead
> > and will slow the server unacceptably.  We need to keep local copies,
> 
> Yes.  I cannot imagine a validating production operation of any size
> that should not want local copies.

Precisely.

> > which means we're stuffed if you change the DTDs by stealth (i.e. without
> > creating a new public identifier for the updated DTDs).
> 
> An http HEAD request usually receives a header "Last-Modified".  As
> long as the http server provides that, there is no added value in
> changing either public or system identifier for a glitch-level error
> correction.

No need for that.  GET with the If-modified-since header will do
the job more simply.  But to do that at run time is still a big hit,
due to the sheer number of modules in the MathML DTD (among other things).

> Gnu "wget" is timestamp sensitive through ftp and http according to
> its docs.  (I've never given it much attention.)

Scripting it is pretty trivial.  The hard part of it is compiling the
list of URLs to check, and cross-referencing them to the PUBLIC ids.

-- 
Nick Kew