We might be destroyed while building an index.

This alters the ctags plugin design a bit to integrate better with the new
libide-core design. We no longer use a service, but simply add an object
onto the IdeContext object graph. We also use IdeNotification to propagate
information about the process to the UI which is monitoring notifications.

Now that g_clear_pointer() has type checks, we want to use that instead of
our wrapper from dazzle.h.

Anything we can do to reduce the priority here.

It would be nice if we could cgroup things, but we don't really have that
level of control.

Unsure of what we can do on FreeBSD, but I presume that nice will work
too (if someone can verify).

This uses the new dzl_clear_pointer() to free data instead of
g_clear_pointer() because it allows us better type checking of
the callback parameter and the data to be freed.

We want to keep using this going forward so that we catch errors
sooner from miss-use of incorrect callbacks.

Our templates have moved away from this, and we can remove it from our
sources too.

This doesn't yet port the service to IdeTask, as that is more complex due
to using DzlTaskCache.

This tries harder to avoid ignored directories as determined
by the VCS.

This uses the cache dir for the project so that reaping files
at purge time is simpler.

As the project has grown, we've let things get a bit out of their
logical place. This does a bit of cleanup and tries to bring some
of the embedded resources closer to where they are used.

But more importantly, this allows us to clean some things up to
the point that we can move forward statically linking a bunch of
the plugins into libide. The plan here is to speed up the initial
loading by avoiding lots of disk I/O stats which are currently
hurting us.

Mohammed Sadiq authored May 13, 2017 and Christian Hergert committed May 17, 2017

https://bugzilla.gnome.org/show_bug.cgi?id=782580

This line was forbidden us to use the "ctag-path" setting

This changes how we do ctags indexes to be a bit more friendly on larger
codebases. Now, we do a single index pass at startup of the project to
update all our ctags indexes. Then, when a file is saved, we only reindex
that directory, ignoring the other directories.

When reloading the indexes, we only pop the cache if the mtime is newer
so that old indexes shouldn't get reloaded/replaced.

In practice, if i save a file like src/foo.c, only src/tags will get
reloaded. Note that src/tags is actually based on the tags directory
in ~/.cache/gnome-builder/tags/$project_id, not the actual source tree.

I think there is a bunch more we can still improve, but this should make
things a bit better.

We no longer need to run on the host as we can bundle ctags in the bundle
to ensure we have access to it.

We don't bundle ctags in the flatpak version, so ensure we are running
ctags on the host system.

We want to use the same subprocess launching that we use elsewhere for
consistency and potential features we might need.

We wan't persistent storage for TMPDIR when generating tags because
it can result in large files being created (which can exhaust ram if
/tmp is tmpfs).

This should also be safe on BSD since we are just using ~/.cache/...

The libide/ directory was getting a bit out of control. The number of files
in the directory was simply too much to make it easy for newcomers to
find what they are looking for.

This breaks things up into a bunch of, somewhat logical, directories.

We no longer use a bunch of crazy include paths with -I, and instead
include the directory name when including as such:

  #include "buffers/ide-buffer.h"

This will hopefully make it easier to navigate, and basically the same to
open files if you use fuzzy search to open files.

Megh Parikh authored Feb 18, 2016 and Christian Hergert committed Feb 18, 2016

This would avoid cluttering of the cache directory with project related stuff

https://bugzilla.gnome.org/show_bug.cgi?id=761391

Right now global variables start with a g and use
camel case.

Those conventions are inconsistent with the rest of the
coding style which uses underscores and no hungarian
prefixes.

This commit does a giant sed over the codebase with some
manual tweaks to fix it up.

If the build system implements IdeTagsBuilder, we can defer our building
of tags to that instead of doing it ourself.

This commit moves many features from libide which were provided by
extension points to instead be plugins. Doing so required changing
how we perform our dependency injection. In particular, we want to
be able to reload extensions when plugins are enabled/disabled.

This is provided by IdeExtensionAdapter and IdeExtensionSetAdapter.

IdeExtensionAdapter provides access to an extension that may be
reloaded or changed during plugin loading based on a match key
and value. The key should be an "X-Foo" style external data key
in a libpeas plugin. The value is the value that should be matched.
The key's value in the plugin file can have multiple values specified
separated by a comma (,).

This patch also lets us remove IdeLanguage, as it becomes fairly
useless now that we can load things via plugins with match keys.
Longer term, I'd like to cleanup IdeFile to subclass GtkSourceFile.

I've started to put language specific stuff into "language pack"
plugins.

https://bugzilla.gnome.org/show_bug.cgi?id=750002