Thursday, December 17, 2009

The Quest Never Ends

In my last blog post, I alluded to our quest to make Pidgin perfect (in our own eyes, at least). This is a quest that by its very definition can never end, because as we near "perfection," there will always be something else that crops up to demonstrate we haven't reached perfection quite yet. This "quest" can have elements of many different forms. A few months ago, one of those forms was discussed on our development mailing list.

As a good many people are aware, Pidgin has long supported GTK+ and GLib 2.0.0. Several years ago when GTK+ 2.0.0 first came out, we underwent a nine-month rewrite of our user interface (UI) to move from the older GTK+ 1.2.10 (or newer) to the new GTK+ 2.0.0. Since that time, until 2.6.0, we've always maintained compatibility with GTK+ 2.0.0 through the use of conditional code that disabled certain features or UI elements that required newer versions of GTK+ or GLib than what was available at compile time. In some cases, we had conditional use of code to work around the lack of certain convenience functions present in newer versions of GTK+ or GLib. In still other cases, we actually carried (that is, distributed in our source tarballs and compiled where necessary) the source of several GTK+ widgets in order to make our UI work for users with older versions of GTK+. Over the years, this has become more cumbersome, to the point that for 2.6.x, it became impractical to maintain compatibility with GTK+ 2.0.0--when we realized this, we changed our requirements to GTK+ and GLib 2.4.0, which contained the features we needed.

The increased difficulty of supporting older libraries prompted me to bring a discussion up on our development list prior to the release of 2.6.0 asking for a vote, discussion, etc. on raising minimum GLib and GTK+ version requirements for Pidgin 2.7.0. This discussion has come up before and been shot down. This time, I put it to a vote and gave quite a while for votes to be cast. There were no "No" votes cast. The versions we voted on were GLib 2.12.0 and GTK+ 2.10.0. These versions allow us to support reasonably recent Linux and UNIX systems, as those GTK+ and GLib versions were 3 years old at the time of the vote, while also making our lives significantly easier since we no longer have to care about really old versions of GTK+ or GLib. Ideally, I would have liked to have GLib 2.14.0 and GTK+ 2.12.0 as the minimums, but I was trying to reach a middle ground that would avoid angering too many people.

This change in the minimum required GTK+ and GLib versions has a few consequences. Obviously, Linux and UNIX distributions that don't ship GTK+ 2.10.0 or newer won't be able to support Pidgin 2.7.0 when we release it. On these systems, users can, however, compile GTK+, GLib, and friends, then compile Pidgin. This is really a lot of work, but some users may be willing to go through it.

For our Windows users, there will be some major changes. When building and linking our releases, we'll be stepping up to GLib 2.18.0 and GTK+ 2.14.0. We already ship this version or newer in our installers, but now we'll actually be linking against these versions. The consequences of this are that Windows NT 4.0, Windows 98, and Windows ME will no longer be supported. These newer versions of GTK+ and GLib require features that just aren't present in those old operating systems. None of these operating systems have been commercially supported (by this I mean modern games, word processors, spreadsheets, etc.) for years, and even projects like Firefox have stopped supporting them.

Additionally, our Windows expert, Daniel, has made some changes to the Pidgin installer and to our crash report generation for 2.7.0:
  • Instead of installing GTK+ in a system-wide configuration, we will be changing to installing GTK+ local to Pidgin. This means that it should be harder for us to conflict with other GTK+ applications on Windows. This has long been requested, but we just finally got around to doing it.
  • Debug symbols can now be read from parallel copies of files. Normal installations of Pidgin ship "stripped" binaries--that is, there is no information useful for generating crash reports. Now, instead of replacing all the existing files, the installer will offer the option to install debug symbols. Selecting this option will install parallel, unstripped copies of every file with the extension ".dbgsym" to a special location.
  • The installer will allow choosing which Pidgin translations to install.
  • The installer will have "online" and "offline" variants. The "online" variant will include only Pidgin. GTK+ and debug symbols will be downloaded as needed. The "offline" variant will include both GTK+ and the debug symbols.
Also among the changes coming up for Pidgin 2.7.0 is the switch from the eggtrayicon widget we have carried in our source tarball for years to the GtkStatusIcon implementation that was added in GTK+ 2.10.0. There are still a few bugs to work out with this change, but I believe once we have those ironed out, our notification area icon will behave better for a number of users who have been experiencing difficulties.

Of course, these changes aren't the only ones that will make it into 2.7.0. I have my own plans for a few additions, plus I won't allow 2.7.0 to be released without a few other features being merged in. But so far, it looks like Pidgin 2.7.0 will make some long-awaited forward strides. Hopefully everyone enjoys it!

Also, to those who celebrate the holiday, Merry Christmas!