[pacman-dev] Improving -Qi/-Si code

[Allan McRae]

On 30/12/13 15:06, Simon Gomizelj wrote:
> We actually do not need to maintain two tables. I've merged it with
> pacman currently and I don't take that approach. I only did it with
> alpm-dump because it was quicker to just dump out statically initialized
> tables for testing. Was probably a poor choice for a demonstration.
> 
> I have a table building function instead, along the lines of:
> 
>     void build_table(char **table, alpm_pkgfrom_t from)
>     {
>         memset(table, 0, sizeof(char *) * LAST_ENTRY);
> 
>         table[ENTRY_REPOSITORY] = _("Repository");
>         table[ENTRY_NAME] = _("Name");
>         table[ENTRY_VERSION] = _("Version");
>         table[ENTRY_DESCRIPTION] = _("Description");
>         table[ENTRY_ARCHITECTURE] = _("Architecture");
>         table[ENTRY_URL] = _("URL");
>         // snip
> 
>         if(from == ALPM_PKG_FROM_SYNCDB || config->op_s_info > 1) {
>             table[ENTRY_REQUIRED] = _("Required By");
>             table[ENTRY_OPTIONAL_FOR] = _("Optional For");
>         }
> 
>         // more branches
>     }
> 
> So rather, the actual effective change, as far as design goes, is this
> allows us to break dump_pkg_full into two discrete steps rather than all
> at once; figure out what to print before you start printing.
>

OK - that sounds good to me.

> In fact, all this table building stuff can entirely stay inside
> package.c if we change dump_pkg_full to expect an alpm_list_t * instead
> of a alpm_pkg_t *. It doesn't need to be exposed to sync.c/query.c (I
> think, I'm not fully confident I've integrated the new system
> everywhere).

I'd have no objects provided it makes sense to always pass a list (I
would need to look at what pacman is passed in single package case).
Otherwise you could add dump_pkg_list() and call the
dump_pkg_{list,full} as needed.

I await the final patch submission.

Allan