use --modes=best to avoid errors from not having a particular kind of
stream available. (I'm not at all clear where the errors come from: I
have never needed this before, but some users do).
Well this is horrible. Now that execute-extended-command is in Lisp,
using completing read, (this-command-keys-vector) ends up returing
[13] for a function executed using M-x iplayer RET (rather than the
more useful whole extended command key sequence). This means that
storing the key sequence for the last command and trying to execute it
later mostly has the effect of adding linebreaks. (The old code works
if the command was invoked through a key sequence rather than through
extended command; I think this suggests that execute-extended-command
should be modified to update the various key variables/functions
specially, like it already has been for this-command, but that is not
helpful right now).
So, as well, store the function name and funcall that. Oh, but wait,
some of the functions dispatch on the key sequence used to invoke
them (presets). So we need to pass the key sequence we were invoked
with back. That means we need to have a common signature for
iplayer-functions. This is all so horrible.
Sure, I only notice this when I have to restart my emacs. But once is
annoying enough. The problem was that running M-x iplayer would cause
an error from within execute-extended-command, within the process
sentinel. After trying to debug this and generating a nifty
unbreakable infinite loop, I finally twigged that it was the
frame/window/buffer frobbing that was the problem. Piling hack upon
hack, the answer is that we only want to do the frobbing once before
the first command in the command keys, and once after the execution
has finished. Make it so.
rework commands in light of long cache refresh time
The issue is that of get-iplayer itself managing its cache: if at any
point it decides that the cache is sufficiently stale, it goes off and
refreshes everything, which seems to take a long time -- if I'm
unlucky, breaking my ERC sessions (and in any case making the rest of
the editor session unusable).
Construct a background process and sentinel to manage the cache
updating before calling the real get-iplayer command. The tricky bit
here is to make sure that the sentinel and real body get executed in
the right context. Many things in the world are simpler with lexical
scoping; this would have been one of them, but I am still on emacs 23
and lexical binding is a thing of the future, so instead of making a
bunch of thunks to be called later, we have to save commands (well,
key sequences), but then those key sequences are divorced from the
window and frame that they were originally typed into, so we have to
get that context back. Naive ways of doing that turn out not to work,
because execute-kbd-macro runs its own command loop, which first of
all selects a current frame and window, so if we've navigated away
from the *iplayer* window/buffer we end up adding keystrokes to
e.g. an ERC buffer, which is suboptimal; however, (ab)using the pre
and post command hooks seems to make things work. So do that, and all
is well.
Very raw. Uses get_iplayer (spelled "get-iplayer" on my system); "parses"
the tree output; uses Orgstruct for presentation, and text properties for
user interface. Uncached calls to get-iplayer (the first in any four hours,
unless the user has a cron job) blocks the whole of emacs for a good little
while as get-iplayer spends its time pulling down metadata, and there is
almost no useful feedback when downloading.
On the other hand, it's easier for me than the command-line interface.
projects / iplayer-el.git / log