Don't delay Firefox startup

This fixes https://issues.adblockplus.org/ticket/117. The critical part is really dealing with potential reentrance. Originally I wanted to add reentrance protection to INIParser but that would allow the "done" notification to happen before all lines are processed. So it seems that preventing reentrance in IO.readFromFile() is the best option.

[Felix Dahlke, 2014-03-14 16:10:45]

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
File lib/io.js (right):

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:92: let index = (processing ? -1 : Math.max(data.lastIndexOf("\n"),
data.lastIndexOf("\r")));
This doesn't really seem reentrant to me. What if "processing" is false here,
but actually gets set to true before we try to set it to true below? I think we
need some locking while the variable is tested and set.

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:95: // Protect against reentrance in case the listener processes
events.
"processes multiple events" I suppose?

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:100: buffer = data.substr(index + 1);
Is reference assignment atomic in Gecko? Otherwise we'll probably want some
locking. Same goes for loadEvent and errorEvent.

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:120: onLoad(loadEvent);
onLoad(null) doesn't seem to be what's desired here. Same for onError() below.

[Wladimir Palant, 2014-03-14 18:04:50]

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
File lib/io.js (right):

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:92: let index = (processing ? -1 : Math.max(data.lastIndexOf("\n"),
data.lastIndexOf("\r")));
On 2014/03/14 16:10:45, Felix H. Dahlke wrote:
> This doesn't really seem reentrant to me. What if "processing" is false here,
> but actually gets set to true before we try to set it to true below?

JavaScript is a single-threaded language, reentrance cannot happen at a random
time. A call to Utils.yield() will spin the event loop however, which means that
events waiting to be processed will be processed. That call will return only
once all the events have been removed from the queue.

One such event can be XMLHttpRequest's progress event. So while we are in the
call to listener.process(...) this event handler can be called again (or the
handler for the "load" or "error" events) if the listener calls Utils.yield().

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:95: // Protect against reentrance in case the listener processes
events.
On 2014/03/14 16:10:45, Felix H. Dahlke wrote:
> "processes multiple events" I suppose?

No, "processes events" = "spins the event loop"

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:100: buffer = data.substr(index + 1);
On 2014/03/14 16:10:45, Felix H. Dahlke wrote:
> Is reference assignment atomic in Gecko? Otherwise we'll probably want some
> locking. Same goes for loadEvent and errorEvent.

As I said, the only place where we might get reentrance is the call to the
listener.

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:120: onLoad(loadEvent);
On 2014/03/14 16:10:45, Felix H. Dahlke wrote:
> onLoad(null) doesn't seem to be what's desired here. Same for onError() below.

True, we need to save the original event before nulling it out. Nulling out has
to happen before the call however, just in case the load listener calls
Utils.yield() as well and we get another "progress" event in the meantime (a
rather theoretical situation).

[Wladimir Palant, 2014-03-14 18:08:43]

Fixed wrong callback parameters

[Felix Dahlke, 2014-03-14 21:32:49]

LGTM

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
File lib/io.js (right):

http://codereview.adblockplus.org/5731880649359360/diff/5629499534213120/lib/...
lib/io.js:92: let index = (processing ? -1 : Math.max(data.lastIndexOf("\n"),
data.lastIndexOf("\r")));
On 2014/03/14 18:04:51, Wladimir Palant wrote:
> On 2014/03/14 16:10:45, Felix H. Dahlke wrote:
> > This doesn't really seem reentrant to me. What if "processing" is false
here,
> > but actually gets set to true before we try to set it to true below?
> 
> JavaScript is a single-threaded language, reentrance cannot happen at a random
> time. A call to Utils.yield() will spin the event loop however, which means
that
> events waiting to be processed will be processed. That call will return only
> once all the events have been removed from the queue.
> 
> One such event can be XMLHttpRequest's progress event. So while we are in the
> call to listener.process(...) this event handler can be called again (or the
> handler for the "load" or "error" events) if the listener calls Utils.yield().

Ah, thought this was actually proper multithreading. But yeah, currentThread
pretty much implies that it's not. So listener.process() can call Utils.yield()
which can in turn call onProcess() while we're still in the middle of
processing. Got it.