Instead of storing everything in key / value pair hashes, we store as much as possible in an array and then reference the items in the array using metatables.
This provides us with a much lower memory footprint and just one level of table lookup indirection (so I think the speed will not be noticeably different)
It is not 100% clear that we've gotten ALL the memory leaks, but it
seems much much much better and doesn't look like it's going to die
immediately or as often anymore :)
This gets rid of a slightly annoying difference between files that fit
in the preview pane and ones that don't: hitting the scroll-preview
mappings would complain about a closed stream. Always using the pager
ensures there's always a process at the other end, even if asking it
to scroll doesn't actually do anything.
* [WIP]: Mon 28 Sep 2020 01:08:24 PM EDT
* add much much better tracking. so much less hax
* status updates, oneshot job updates, etc.
* remove temp function
* add status function
* asdfasdfasdf
Warn users if they have `report` set to a bad value
In the future, we could try and change it and put it back, but that
feels a bit bad. If anyone can tell me WHY you would want this, then we
can do it.
* feature: Vim help-tags picker
* fix: filtered results were wrong because of missing `entry.value`
* fix: filtered (Vim only help) items are listed in results as empty
entries.
* fix: avoid search history pollution by replacing / in cmd returned by taglist() with search()
* fix: improve search() formatting
* fix: escape tilde in search() command
* fix: improve help-preview
* fix: improve search()
* fix: search() string fixes.
* fix: use no magic to do magic
Co-authored-by: TJ DeVries <devries.timothyj@gmail.com>
* Don't execute `command` if target buffer is already loaded in window.
* Print node_text before entry.kind in ordinal string.
* Add new builtin for listing and reloading Lua modules.
* Revert "Add new builtin for listing and reloading Lua modules."
This reverts commit 65d9d0294eae92b347b38b8dcf807fbcdd5783fe.
This should reduce memory usage at least for large jobs. We won't save
all the results into the job. I _think_ this will be half memory usage
for many situations.
