On 2018-02-23 02:52 PM, Lex Trotman wrote:
[...]
As I have been saying for a while, even for existing plugins I would make the GTK3 one a separate plugin [...]
This isn't a good idea because it causes way too much unnecessary maintenance burden. To save some #ifdef in a single file to smooth over the few differences between GTK+ 2 and 3, you're recommending to duplicate tens of thousands of lines of plugin code. This not only includes duplicating all of the C code, but also all of the build system code, translations, docs, etc.
With a #ifdefed plugin it takes effort to build it to support both
It's less effort to maintain a compatibility file/module with the few differences rather than having loads of duplicated code/infrastructure. It causes less burden on contributors and maintainers to edit a single code base with a few #ifdef stuff than to try and port changes between diverged plugins, or to edit one or the other, further diverging the forks.
To give an example, PR #697 ports Scope plugin from GTK+2 to GTK+3 and only touches +57,-26 lines in 7 files. That's roughly the number of lines which diverge between GTK+2 and GTK+3 for the plugin, of a total of around 12,000 total lines of code in 44 files (estimate of only C code using cloc utility). That's around 1% of code which differs between GTK+2 and GTK+3; IMO not enough to warrant a fork/duplicate.
and again to untangle it when GTK2 goes away, with two plugins you just delete one
Going the route of isolating the differences into a single compatibility header/module makes it even easier to drop GTK+2 support. For example in Geany core C code, we only have to delete the one gtkcompat.h header and GTK+2 support is gone. There is surely some other #ifdef stuff mixed into other parts of the code, but it's far less work to delete it later than the amount of work to maintain separate forks for GTK+2/3 when 99% of the code is in common.
Regards, Matthew Brush