On iOS, the main thread does framebuffer operations just before it requests a rerender of the view contents. This means that GL operations on the Map thread will interfere with that and cause a crash. While most GL operations happen throughout rendering (which is synchronized with the main thread), deletion of shared_ptrs release in turn their OpenGL objects. This may happen at any time and isn't synced with rendering.
This patch changes this so instead of deleting all objects immediately, we are collecting all abandoned IDs and delete them in bulk once the main thread has given us permission to use OpenGL calls (i.e. during render()).
When using Valgrind for debugging memory leaks and access violations, we
often see a lot of false positives, specially when dealing with OpenGL
that does it own buffer management. This file is an attempt to filter
these false positives.
Note that we are only filtering for the open source Radeon driver and
the list will probably grow to accommodate other setups.
To use the Valgrind with the suppression file for hunting memory leaks,
try something like:
valgrind --leak-check=full --track-origins=yes --suppressions=[path/to/valgrind.sup] ./mapbox-gl
This podspec was originally lifted from mapbox/mapbox-ios-sdk, which was structured and built quite differently. We could have the podspec build this repo as is, using scripts/package_ios.sh, but that would require developers to install the same version of Xcode as Travis is using (see #977). So this podspec instead pulls a prebuilt library. Much simpler.
Also package a LICENSE with the archive to satisfy a CocoaPods warning.
Fixes#829.
As a best practice for the beta and a workaround for #1014.
Also reversed the argument order of the existing init method. Reads better this way, but you’ll likely have to change your initialization to match the new signatures.
To make it easier to debug, add the Environment ID and the Thread name
to the log messages. For instance log messages from the TileParser will look like:
[WARNING] {0}{TileWorker_15/18653/9486}[ParseTile]: some relevant warning message
| | | |
| | | +-> Component
| | +--------------------> Thread name
| +----------------------------------> Environment ID
+------------------------------------------> Severity level
Log messages that are not inside an Environment::Scope will work
normally and will look like:
[WARNING] [JNI]: some relevant warning message
Fixes#882.
We are also removing the code that is passing the Environment around
and instead, we get now the Environment in the TileData using
Environment::Get().
When processing the job, the work will be temporally registered to
the Environment as "TileWorker_0/0/0" which can be used for logging
in the future. At the end of the job, it gets unregistered
automatically from the Environment.
The Environment will now keep track of every thread that is running
within a Environment::Scope. Once the Scope object is created, it will
automatically register the thread to the ThreadInfoStore with some
additional information and it will automatically deregister as the stack
frame rewinds (which can be useful for Threads that are temporally in
use by an Environment like Workers).
The ThreadInfoStore could be using the thread-local storage, but I think
that making the Environment aware of what threads are associated to it
can be useful for writing a mechanism for posting messages between threads.
Withing a Scope, the Environment associated to the thread can be queried
at any time which will save us from juggling a reference to the
Environemnt around. Finally, the type of thread (Main, Map, etc) can
also be queried at any moment to make sure a piece of code is running
on the desired thread.
- OpenGL ES 2 doesn't allow NPOT textures with wrap-around
- The Sprite object reported the map's pixelRatio, even though it loaded @2x assets
- Copying icons from the sprite into the atlas now uses bilinear scaling to scale up to the actual size
We are using X11 symbols directly like XFree() so we need to link
with it explicitly. Previously these symbols were being resolved
somehow by libGL, but apparently they are now hidden - good - in
the custom version we are using since 5e40f72b7.
Fixes#1005.
