Re: Transactions and specfiles...

[Jeff Johnson <jbj redhat com>]

On Fri, Aug 22, 2003 at 08:59:35PM -0400, James Olin Oden wrote:
> On Fri, 22 Aug 2003, James Olin Oden wrote:
> 
> > Hi Jeff or whoever knows,
> > 
> > I was poking around in build.c and I found that in order to parse 
> > a spec file you pass an rpmts to parseSpec(), and it then
> > sets the name of the specfile in the transaction with 
> > rpmtsSpec().  My question is why do you need a transaction to parse
> > a specfile?
> >
> I looked further, and found some of my answer.  Your using the 
> transaction to check build dependencies using the same routine you
> use to check install dependencies (rpmtsCheck()).  That said, there
> is a little jig you do that goes like this:
> 
> 	1) Parse the spec file with parseSpec() passing it the 
> 	   transaction.
> 	2) parseSpec will put a pointer to the newly created spec
> 	   structure in the rpmts via rpmtsSetSpec.
> 	3) In the routine that parseSpec returns to, you then
> 	   call rpmtsSetSpec again setting it to null, but as
> 	   a side affect rpmtsSetSpec returns the pointer to
> 	   the spec structure (created in parseSpec).
> 	
> You then use that pointer to the spec structure to do other things 
> (build a SRPM header in core, and so on).  What I don't get is 
> why not just return the pointer to the spec structure, or have
> parseSpec take a pointer to a spec pointer, and then just set that
> directly?  The way your doing it seems convoluted, or am I missing 
> something (which I have no doubt I am (-;)?
> 

Yup, convoluted mainly for hysterical reasons, recent changes drilled
an rpmts object only where necessary, not completely as on the install
ide, through the rpmbuilkd API.

> On a seperate note, further down in buildForTarget(), you call
> buildSpec() with an rpmts as the first argument, but in this case
> buildSpec() does not use the rpmts at all?  Did you use to use the rpmts,
> decided not to continue doing so, but by then someone was calling 
> buildSpec directly, and removing the rpmts arg would have broken 
> their code?

Yup, starting to drill an rpmts object everywhere even if not (yet) used.

73 de Jeff

-- 
Jeff Johnson	ARS N3NPQ
jbj@redhat.com (jbj@jbj.org)
Chapel Hill, NC