so - some people have issues if we open devices. why... i don't know,
but add an option to toggle this and be conservative and have it off
by default
@fix
so our writes sometimes would get stuck because kernel io buffers are
full and writes are slow. on specific machines with super slow write
media and small amounts of ram this was bad.
this moves writing totally to threads. the eet file is opened in a
thread and closed in the same thread. only the eet_write/eet_data_write are
done in the mainloop. this is a 'walk struct, serialise it and compress
it" which compared to blocking for possibly multiple seconds in a
write/close/rename backup cfg files doing real io to kernel 9even
though kernle should buffer these)... is a hell of a lot better.
so sure. we block lock enough to walk the structures/lists, encode the
blob and put it through a fast lz4 compress cycle and drop into memory.
the actual write happens in the thread when the file is closed and
that is a vast improvement if you hit these cases.
it used ot be a separate process to run to hide e starting in the bg
on a slow hdd loading modules etc. but due to compisitng and other
changes its all internal now, so keep it on always as it guarantees a
better smoother experience with less complexity to maintain.
if theme doesn't support, fade-in will not happen (it'll just appear),
but if it does... e will fade to black for a restart, then fade back
in nice and cleanly.
this means no leaked fd's between restarts too. cleaner. it also
encorces "die with parent" for enlightenment_start too as another
bonus in addition to its own fifo handling for singleton access per uid.
WARNING: you need to log out and log back in since the "protocol
expectations" (what exit codes do what and what parent and child
process are responsible for) changed so to get them both back in sync
you need to log out and in.
so e has had a bit of a mess of setuid root tools to do things that
are essential to maintaing a functioning desktop/laptop/tablet/phone
like device like shutting it down or suspending or sqizzling the cpu
governor or messing with the backlight or... you get the idea.
this has been spread around enlightenment_sys and other special
purpose tools. this now unifies it into a single "always there"
enlightenment_system backend setuid root slave process whose job it is
to do all these things via a stdio protocol in an organized way. this
means latency to do something is lower, but at the expense of
consuming ram and a lurking process. unfortunately the lurking will be
needed soon when i add ddc support to make it even vaguely efficient,
so it's a cost i guess we have to pay now. we'll need this in future
as well for some stats collection and more.
still need to port existing code to use this instead of the existing stuff,
and then remove of the old stuff.
since now we basically bring everything up THEN start rendering once
we hit the main loop there is no point having init splash strings
anymore. it is no longer an external process. until e's compositor is
up you will see nothing anyway. if it is up these days we've finished
init as a smoothing out of startup has move everything before main
loop starts. so this code is all kind of useless, thus remove. cleanup
code.
this moves a bunch of things that were being don after loop starts to
before so they don't interrupt framerate as well as launch a thread
that preloads file(s) - for now just edj theme files so stuff is
pre-cached by the time loop starts later so we don't stall s much
leading to smoother startup/fade in and better framerate. it's really
visible (especially on my rpi3).
we dont need to delay module loading anymore so remove allof this -
priority too shouldn't be necessary as well and was just a workaround.
this remvoes all of that removing some e module api's and module
struct members as well as some global config fields. it also ensures
all config files are up to date as well.
so when e was e.g. shutting down the system it'd sometimes - or all
the time, restart and not shut down. this fixes that. e was being sent
a HUP signal while doing this causing e to go "ooh HUP - i shall
restart" which si what daemons without a controlling tty do with
signals commonly. this isn't what we want from e though in this case,
so delay itby a bit and ignore if we're in the process of logging
out/halting etc.
this really only affects wayland mode.
@fix
Summary:
Moving _xdg_data_dirs_augment call to earlier in
initalisation process. Currently first-run will
leave XDG_MENU_PREFIX broken, thus the application
menu will be empty in E.
Seems eio_init might be the culprit.
Moved to above eio_init.
Test Plan:
* Launch E from X from a console (startx)
* Check application menu (empty).
* Apply patch.
* Launch E from X from a console (startx)
* E's Application menu should be populated.
Related to https://phab.enlightenment.org/D7534
Reviewers: #committers, raster, cedric, zmike, devilhorns
Reviewed By: #committers, zmike
Tags: #enlightenment-git
Differential Revision: https://phab.enlightenment.org/D7535
is e crashes, catches it and restarts while locked you end up
unlocked. this lets enlightenment_start know this lock down state and
it sets an env var to ensure locking happens on restart after recovery.
this seems to happen to @manmower in T6619. it may just be timing
changes that changed the start loop time to be at init of ecore maybe ...
but it was doing a forced hide and not a clean "done"
that animates/fades. so fix this along with deferring show until the
first render pre.
this fixes T6619
This reverts commit 40e8bb044c.
EFL master solves this properly
EFL stable currently has a bug that solves this accidentally
so there's a window of about 3 commits in git where it's a problem,
and nobody should run those commits in production :)
This fix caused some problems when restarting E, as stdin would
cloexec and then the following startup would try to set cloexec
on an invalid fd and generate an err that generated a cri because
of an invalid log domain.
Someone should probably figure out the log domain thing.