RE: patch 20031120

> > 
> > However I changed tds_iconv prototype and some behaviour. 
> io parameter 
> > was replaced with a flags parameter that specify direction, 
> previous 
> > error and chunking behaviour. tds_iconv return a set of errors, not 
> > only a single error.
> > Onestly I don't like that much the change (perhaps it's 
> best to add a
> > int* flags parameter??)... however it works.
> 
> Ugh, I don't like it either.  I'll review your mail in case 
> you told me already, but I don't understand why tds_iconv 
> should care about incomplete end-of-buffer sequences.  It 
> should return normally, leaving errno set to EILSEQ, and let 
> the caller determine if that's an error warranting a user 
> message or not.  
> 

In chunked conversions every single chunk can raise EILSEQ so I return
error that tds_iconv had and I pass them back to next tds_iconv so
tds_iconv know if it has already sent a message. Also in chunk
conversions tds_iconv do not send a message on EINVAL.

> Please, let's not overload TDS_ICONV_DIRECTION.  We can bury 
> a flag in TDSSOCKET or TDSICONVINFO if need be.  
> 

It's better in TDSICONVINFO but caller should send a "this is the first
chunk" flag... perhaps this can be a better solution... add a field to
TDSICONV to store old errors and caller should say if this is the first
and/or the last chunk (in no-chunk conversions it's always the first and
the last chunk). So you can also reset iconv state (iconv(cd, NULL,
NULL, NULL, NULL) )... Does this sound reasonable ?

freddy77