makepkg --source should be used to upload packages.
It provides a bit of error checking and it's good to support only a
single format here.
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
Add msearch type to the list on http://aur.archlinux.org/rpc.php
Signed-off-by: Laszlo Papp <djszapi@archlinux.us>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
changed file name, as the 2 is relatively meaningless.
I think at one time it qualified it as a rss2.0 feed, but the
distinction is valueless at this point, since the generated xml
itself declares the rss version.
make sure to set an apache rewrite rule for any production systems
that require it.
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
jsonp callback should have a content type of text/javascipt,
since it specifies a callback function wrapping json data,
and is not soley json data.
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
Currently AUR does not check the uploaded packages, whether they are
correctly in UTF8 encoding. If there are fields (such as $pkgdesc)
that contain such non-UTF8 characters, the rpc interface chokes on
those fields: even if there's data successfully retrieved from the
database, rpc returns "null".
In an effort to make such errors in the PKGBUILDs debugged more easily,
let's force rpc to have more useful output: try to convert non-UTF8
fields to UTF8. If it's a success, good. Partial success (some characters
showing as \uXXXX) is ok. Failure results in error message printed to that
field.
Signed-off-by: Gergely Imreh <imrehg@gmail.com>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
All of these are sourcing function libraries so we don't need to include
them more than once. Things that insert actual HTML into the output were
left calling include().
Signed-off-by: Dan McGee <dan@archlinux.org>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
comments need to be removed before concatenating lines, otherwise
not matched brackets can cause problems on submit
Signed-off-by: Gergely Imreh <imrehg@gmail.com>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
Refactor the query to use explicit LEFT JOINs, which appears to be handled
by MySQL in a saner fashion than the previous implicit join syntax. This is
part two in a slight fixup for observed slow queries in the production
environment. With the new indexes and this fixup, a particular iteration of
this query will examine only 13346 rows instead of 272060.
Signed-off-by: Dan McGee <dan@archlinux.org>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
To put a long story short, when we do joins on these tables in our
pkg_search_page() function, we always join on both the user ID and package
ID columns. By creating multicolumn indices, we can always get the exact row
we are looking for in the table.
The benefits of adding a unique index should also speak for themselves, as
we previously did not have this on either of these tables.
This is part one of a two-part series to address the fact that this query
was often showing up in our slow query logs.
Signed-off-by: Dan McGee <dan@archlinux.org>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
It's performance improvement day today. For non-superusers, we were hitting
the database twice per comment on a package- once to get the UID, and once
to check the owner of the comment. The best part is we already knew the
owner of the comment, and we only need to get our own UID once.
For viewing a package like yaourt, this cuts a single pageview from over 700
queries to around 18, which is still not great but a pretty big improvement.
Signed-off-by: Dan McGee <dan@archlinux.org>
Signed-off-by: Loui Chang <louipc.ist@gmail.com>
