You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Aug 8, 2023. It is now read-only.
Repository navigation
This repository was archived by the owner on Aug 8, 2023. It is now read-only.
The stack trace from #6883 indicates that the MBGL_CHECK_ERROR macro was compiled to the error-checking variant in a release build. Most likely, this means that release builds are neglecting to define NDEBUG, and therefore including code intended only for debug builds. This includes GL error checking, debug level logging, and asserts.
@kkaefer, is this something that may have regressed in the switch to cmake?
Nope, Release builds are still built with NDEBUG. However, it seems like f41546e changed the default BUILDTYPE for make ipackage to Debug. Apparently this was in response to 9732f02 defaulting to Debug in the rest of the code as well.
When I build locally with SYMBOLS=YES BUILDTYPE=Release make iframework and SYMBOLS=YES make iframework things work as expected in the repro app where there is a zero rect framed map view (as expected means log spew in release and crash in debug). So it does seem like something unexpected happened when I published beta 2. We don't intend to make it possible to publish a debug build using the publish script so I don't think I "ran the script incorrectly" so I will keep digging to figure out what happened.
When I built beta.2 I used Xcode 7.3.1 because #6842 (comment). I'm not exactly sure why, but something about xcode-selecting that toolchain (I was on Xcode 8 just before that) seems to have caused this Debug output even though the scripts should always result in Release output.
I spent a bunch of time on this before deciding to turn my computer off and turn it back on it again. This appears to have "fixed" the problem. I'm going to close this and will keep an eye on future releases. I think the solution for this already ticketed: #2844
Activity
kkaefer commentedon Nov 3, 2016
Nope, Release builds are still built with
NDEBUG. However, it seems like f41546e changed the defaultBUILDTYPEformake ipackagetoDebug. Apparently this was in response to 9732f02 defaulting toDebugin the rest of the code as well.@friedbunny @incanus @1ec5 did you accidentally build iOS packages without
BUILDTYPE=Release?kkaefer commentedon Nov 3, 2016
We could add something like this:
mikemorris commentedon Nov 3, 2016
We have a hard fail when trying to publish a Node.js package with
BUIILDTYPE=Debugon Travis or Bitrise - would something like this be helpful?boundsj commentedon Nov 3, 2016
@kkaefer we don't use the default when we publish a build. We use a script and it sets the build type to release. See https://github.com/mapbox/mapbox-gl-native/blob/release-ios-v3.4.0/platform/ios/scripts/deploy-packages.sh#L53
1ec5 commentedon Nov 4, 2016
One theory is that we’re building the SDK itself in Release, but it’s linking to an mbgl static library built in Debug.
boundsj commentedon Nov 4, 2016
When I build locally with
SYMBOLS=YES BUILDTYPE=Release make iframeworkandSYMBOLS=YES make iframeworkthings work as expected in the repro app where there is a zero rect framed map view (as expected means log spew in release and crash in debug). So it does seem like something unexpected happened when I published beta 2. We don't intend to make it possible to publish a debug build using the publish script so I don't think I "ran the script incorrectly" so I will keep digging to figure out what happened.boundsj commentedon Nov 8, 2016
When I built beta.2 I used Xcode 7.3.1 because #6842 (comment). I'm not exactly sure why, but something about
xcode-selecting that toolchain (I was on Xcode 8 just before that) seems to have caused this Debug output even though the scripts should always result in Release output.I spent a bunch of time on this before deciding to turn my computer off and turn it back on it again. This appears to have "fixed" the problem. I'm going to close this and will keep an eye on future releases. I think the solution for this already ticketed: #2844