FreeTDS Development Group

["James K. Lowden" <jklowden AT freetds.org>]

ZIGLIO, Frediano, VF-IT wrote:
> > Instead, let's have a release-by-release scheme.  
> > 
> >     ChangeLog:      changes in CVS HEAD since the last release
> >     ChangeLog-0.82: changes since 0.64
> >     ChangeLog-0.64: changes since 0.63
> 
> I mainly split between years cause committing ChangeLog every time was
> starting to be a nightmare...

I know. :-)

> I don't know if automake require a ChangeLog but I think we can put a
> "See ChangeLog-xxxx" line...

Automake doesn't care.  

January isn't a significant marker.  Releases are.  If ChangeLog has only
changes since the last release, and it's taking too long to commit, that's
a hint that it's time to release!  ;-)

> > Good ideas all, except I want to continue to support BSD 
> > make.  I don't see any reason not to.  
> >  
> 
> Is that I don't know if BSD make have some extension that could make
> no-recursive make easy to develop.

Let's try, OK?  BSD make has some different ideas about how to do things,
but so far we've been able to find a useful intersection of features.  

> I found a good way to have customizations within directory. Using GNU
> make you can have things like
> 
> src/odbc/unittests/% : CPPFLAGS += -I$(ODBC_INC)
> 
> that is: if target is in directory src/odbc/unittests add -I$(ODBC_INC)
> to CPPFLAGS !! Simple and working!!

I need to look a "man make".  

> > In db-lib -- I don't know if ODBC has a similar feature -- data can be
> > fetched with bound columns (dbbind) or not (dbdata).  For 
> > bound columns,
> > the client allocates a buffer for the data.  When libtds 
> > fetches a row, it
> > needs to allocate buffers only for unbound columns.  
> > 
> 
> Mmm... a question... does dbdata work if dbbind is used?? If yes you
> always have to buffer result...

dbdata works regardless of dbbind.  But: the buffer pointer returned by
dbdata can be the same one provided by dbbind!  

Unfortunately, if the dbbind buffer is too small, dbdata will need its own
buffer and a copy will have to be made to dbbind's buffer.  

I haven't tests dbdata with BLOBs in Microsoft's library.  db-lib has
dbreadtext, which is supposed to read a BLOB in client-defined chunks. 
That would seem to imply repeated calls to libtds to read data from the
server socket.  What happens if you call dbdata partway through?  

The robust way to do this would be to read the BLOB data to a temporary
file and spoonfeed it to the client as requested.  If dbdata is called,
allocate on the fly for that one row only (or use mmap and say it doesn't
work on Win32).  

> > >  That will
> > > avoid copying kernel->libtds->client.  

> Yes, usually referred as zero-copy :)

See?  I'm catching on! 

Regards, 

--jkl