GitLab

At worst it will use a lot of CPU but this will allow to detect buggy
access/access_demux module.

For a start, nobody uses UTF-32 for transmission/storage.
Then, we anyway have iconv() support in the subtitle support (where it
belongs IMHO), if you really wanted to use UTF-32.

The record access_filter will become useless as soon as all demuxers that can
support it will be modified.
The record support is half done. I will also add es_out_t record functionnalities,
and a way to selected one of them if both are supported.

Don't use "p_block->i_buffer = read(...)", i_buffer is unsigned so you
won't catch read errors and will end up crashing using a 4GB buffer
instead. This fixes a DVB segv I was seeing.

Signed-off-by: Pierre d'Herbemont <pdherbemont@videolan.org>

requested one. (Breaking at least mapped file input).

Fix two potential segfaults.
Cosmetics.

This reverts commit 0bcda15e.

It is much more efficient regarding latency as it doesn't bufferize more than needed, and let the module access take care of that eventually.

Enable with --use-stream-immediate. We may want to default it.

Note: --use-stream-immediate will be only effective on access that don't provide pf_block() for now. This is because I didn't benchmark against the Block method.

Here the gain that I did measure is about 200ms (less latency) when using the http access on a loopback.

Reported by Coverity scan : CID 142

bl4 authored May 24, 2008 and Rafaël Carré committed May 26, 2008

Signed-off-by: Rafaël Carré <funman@videolan.org>

...and use it for stats and the config lock

This improves readability a lot and prevent us from reinventing the wheel.

stream: No use to vlc_object_yield() and object that create ourslef, and that a race could possibly corrupt in case of bad usage.

Because there is no proper convention, there is no reason why the object couldn't be detached. We would need to vlc_object_find the input to be sure there is no problem. Let's assume for now that the input creates the access and is responsible for it.

The input already does the job.
Signed-off-by: Rémi Denis-Courmont <rem@videolan.org>

misc/objects.c: Don't rely on vlc_object_destroy() to destroy objects, but expects vlc_object_release to do it when the refcount goes to zero.
* Meaning, that when created objects gets a refcount to 1.
* Destroying is instantaneous and we don't have to poll for a few secondss or so to wait until the object's refcount reach 0.
* We now track vlc_object_t's mem leaks when libvlc_global is released (Hard error for now, so they don't get unoticed)
* We fail hard if an object is released with a refcount of 1 without being detached from its parent and its children, to make sure such cases don't go unoticed.
(make test or make check still pass after that one. VLC is known to leak one object when no module is loaded, this must be fixed).