Undocumented option "bash_hack" is gone.
New function syntastic#util#system(), that sets up the environment and
is now used consistently everywhere.
Unrelated fix for the elixir checker (syntax).
The new option tells the checket to load the file in memory and look for
a comment line "# coding: <codec>". If found, the named codec is used
to decode the file before checking. This makes sense only if said codec
doesn't change the line numbers of the initial file. An example of such
codec is "spec" provided by the "Nose of Yeti" library:
https://github.com/delfick/nose-of-yeti
On terms that support smm / rmm (f.i. xterm), initializing readline
prints the smm sequence, regardless of whether stdout is a terminal or
not, which in turn can make checkers' output unparseable.
Workaround: set TERM to dumb before calling the checkers.
Error sorting can't be done as a postprocess function called from
SyntasticMake(), since the final values of some relevant fields (f.i.
type) might not be known yet at that point. Solution: move sorting to
getLocListRaw(), after per-checker quiet_messages. New checker methods
getWantSort() / setWantSort() are needed.
Second problem: some checkers return screen columns mixed with byte
indices. Solution: compute screen columns as needed. Sadly, everything
about working with screen columns is fragile.
Optimisation: merge handling of global and per-checker filters.
Feature: allow [] and '' values for quiet_messages filter elements.
Feature: allow overrides for quiet_messages filters.
Feature: buffer-local auto_jump and quiet_messages.
Safety: add magic specifiers to catch regexps.
Cleanup: minor restructuring for the forthcoming foreign checkers
feature.
