A major issue is that we currently can't pass any windows' mode on to
the segment rendering method, so non-current windows don't get
highlighted correctly, and segments don't get removed if they have 'nc'
in their exclude_modes setting, and statuslines in non-current windows
don't get resized until the window is focused again.
This class loads all segments from the theme, initializes the highlight
group, assigns all necessary properties based on the JSON theme
configuration, etc. By doing this we basically move the mksegment()
functionality into the theme loader, so this function can be removed at
the cost of making it more complicated to use Powerline without its
theme functionality.
Hard dividers (for segments with different background colors) are now
always drawn, regardless of the draw_divider option, because it looks
horrible when segments with different bg colors don't have a divider
between them.
This resolves an issue with the filename and modified flag
in the default theme where the filename segment had to set the
draw_divider flag based on the contents of the modified flag to be
rendered correctly.
Another rendering pass is necessary before calculating the filler
segment's lengths, because center segments may lose their separators
after removing low-priority segments.
Some unicode and variable assignment issues has been resolved so
everything renders correctly.
This change removes the Segment class as this takes forever to remove
from the segment array when removing low-priority segments. It has
instead been replaced by a wrapper function that works the same and
returns a working dict of all segment properties.
The regex substitution bottleneck in the vim example has been fixed by
using a single-character percent placeholder in vim segments which is
later replaced with a double percent using str.replace().
This commit almost doubles the segment rendering performance. This is
accomplished by caching a lot of data like highlighting groups, moving
some calculations out of loops, and by performing less function calls
overall.
When a width is specified the main speed improvement comes from avoiding
rendering the raw segments over and over until the statusline is short
enough. Instead, the raw rendering is stored as a segment property and
the combined length of all these renderings is used when removing
low-priority segments instead. This results in a maximum of two
rendering passes.
Some "less pythonic" solutions have been chosen some places for
performance reasons, e.g. joining strings instead of appending and
joining lists.
Overall this commit appears to make the performance equal or better than
the legacy vimscript implementation. Later optimizations (in particular
finding another method than remove() for removing low-priority segments)
may make this version of Powerline far superior both in terms of
functionality and performance.
This commit introduces a bunch of changes. Most importantly, it ensures
that split windows and inactive windows work as expected. This is
accomplished with some clever workarounds for vim's statusline
limitations.
Vim statuslines evaluated using the %! statusline item are always
evaluated in the buffer context and not in the statusline's owner
window's context, and this made the previous implementation show the
same statusline for all windows. The only way to ensure that the correct
statusline is rendered to the correct window is to walk through all
windows, and setting the statuslines with a reference to the current
window number every time the current window is changed. This is
accomplished by a set of BufEnter, BufLeave, etc. autocommands.
The Syntastic segment has been temporarily removed due to errors when
referencing the statusline function before Syntastic has been loaded.
When more than one filler segment is specified it may sometimes be
a remainder when calculating the amount of whitespace. divmod() is used
to retrieve the remainder, which is added as whitespace to the first
filler segment.
The examples have been moved into their own directory. The vim example
now uses pyeval() and vim.bindeval() and loads the Python file through
an import (so it gets compiled) for speed improvements.
The segment tree seemed like a good idea at the time, but for the core
code it's probably best to do a single-dimensional array and instead
have some duplicated data (common segment attributes). By removing the
tree structured segments the code is much simpler to understand and use,
and probably quite a bit faster.
If a tree structure is wanted for ease of configuration, it could easily
be supported in the JSON config files for vim statuslines, and then
letting the vim statusline code flatten the configuration into
a single-dimensional array for rendering.
Powerline will probably only be Python 2 in the foreseeable future
because of poor Python 3 support in applications like vim, so this
small change ensures correct unicode behavior in Python 2.
Kim Silkebækken
ZyX