17 Oct
2008
17 Oct
'08
1:39 p.m.
Fons Adriaensen <fons(a)kokkinizita.net> writes:
On Thu, Oct 16, 2008 at 11:40:09AM +0300, Nedko
Arnaudov wrote:
IMHO this should be provided by libjack API itself.
This is same thing that kfoltman requested these days, to be able to
check whether app is started as part of process of restoring lash
session. I'm fine with this and such approach in general.
The trend is to remove usage of argv at all. As
for documentation, I
agree LASH needs better one. In current documentation (somewhat out of
date), argv thing is supposed to be opaque. I.e. LASH app should not
know what lash specific arguments are.
That exactly is the problem. Any app should be allowed
to know. Elementary courtesy towards your client...
Support libraries are not meant to be 'stealthy'. Looking at the source, there's indeed a lot of
deprecated stuff, followed by two flags, and then
the complete argv minus the lash-specific parts.
I don't think blindly copying the complete argv
and giving it back when the client is restarted
later by lash is a good idea.
I have plan to allow user to tweak commandline of both lash first time
launched apps (when launched through lash control app) and restored lash
app processes, before launched. However this is not in my short term
TODO list.
If there is any data in there that is essential
for the client to be restarted later, then the
client would probably include that in its configs
or in its saved config file.
Unless you have app that has no internal state at all. Examples:
fluidjack and jkmeter (ver nice app btw). This simplifies lash
integration for simple commandline apps.
The only exception would be things that are essential
even before the restarted client connects to lash.
But only the client knows which data that would be.
Can you give some example, please?
So it seems it would be a better approach to let
the client supply those parts of its argv that it
wants back the next time, if any, rather than just
taking it all.
Such option seems fine. I'm only opposed to *requiring* such
app<->liblash interaction, because it will make commandline argument
management more tricky.
This issue (and some others, like jack connections)
is not as simple as it seems.
For example, none of my (GUI) apps knows if any
of its configuration comes from compile-time defaults,
/etc/appname.conf, ~/.appnamerc, ~/Xdefaults, or
the command line. They just interrogate the X11
database which combines all of that with the right
priorities (part of libclxclient).
Some of that data may be pointers to application
resources (e.g. the instrument definitions used
by Aeolus) rather than user options. Such resources
may have been moved or replaced next time lash runs
the app. In that case the app wants the current
values, not the old ones. But only the application
knows such things, so blindly copying everything
is not the right way. Since the app is given the
chance to store its config in any way it wants,
it isn't necessary either.
I think the easiest option would be to store internal state in lash and
ignore commandline when restored as part of lash session. Other option
would be to declare such behaviour to lash, either by call of liblash,
or in the desktop entry for your app (X-LASH-XXX fields, we have some of
them already).
--
Nedko Arnaudov <GnuPG KeyID: DE1716B0>