[pacman-dev] [PATCH] Add [ignored] to -Qu output for packages in repos that are not Usage = Upgrade

[Andrew Gregory]

On 01/04/19 at 02:21pm, Allan McRae wrote:
> The behaviour of "pacman -Qu" was very strange...  It would only consider
> packages from repos with Usage = Search (or All), and ignore those with
> Usage = Sync, Install or Upgrade.
> 
> This is because alpm_sync_newversion() used ALPM_DB_USAGE_SEARCH for its
> filtering. Given this function is documented (at least in the source) to
> "Check for new version of pkg in sync repos", I would expect that to look at
> all repos. However, just changing this parameter, would result in a fairly
> silent change in behaviour of this function.  To counter that, add a parameter
> to the function that tells it which databases usage levels to consider.
> 
> Finally, list all available updates in -Qu output, but include [ignored] beside
> those that will not be updated in a -Su operation due to thier repo Usage
> value (in addition to those that are Ignored).
> 
> Fixes FS#59854.
> 
> With thanks to the following who provided initial patches to print [ignored] on
> -Qu operations, which highlighted the larger problem:
> morganamilo <morganamilo at gmail.com>
> Michael Straube <michael.straube at posteo.de>
> 
> Signed-off-by: Allan McRae <allan at archlinux.org>
> ---
> 
> In comments on earlier patches, there was debate about what the
> alpm_sync_newversion() is doing.   Although I expect users of this function
> to likely always pass ALPM_DB_USAGE_ALL, I think changing the function
> signature is the safest way to handle this change.
> 
> Comments?

I was initially on board with this approach, but the more I think
about it, the more convinced I am that having a function that takes an
explicit list of db's filter them based on usage is a fundamentally
bad design.  As I said in the initial discussion, these are low-level
functions that may be used in different contexts, in which case
different usage filters would be appropriate.  For example, this patch
uses newversion with USAGE_ALL because it's being used for purely
informational purposes.  Manjaro's syncfirst patch, however, uses it
to actually prepare an upgrade, making USAGE_UPGRADE the correct
filter.

But, requiring the caller to provide a usage everywhere would defeat
the purpose of making repo usages a back-end feature.  We might as
well just have the caller do the filtering.  Given that there are only
a handful of usage levels, the filtered lists could even be cached,
completely saving us the hassle of iterating over db's that won't get
used.

I think the best way to fulfill the original goal of adding repo
usages would be to move the usage filtering to higher level functions
that have more narrowly defined purposes and take a handle rather than
a db list.  This would allow callers that just want to "do the right
thing" to not have to worry about usages while allowing callers with
more specific needs to use the low-level functions without having to
worry about what extra filtering alpm might be doing behind the
scenes.  It would also result in a cleaner API that doesn't have
functions ignoring databases they were explicitly told to use.