Opened 5 years ago

Closed 5 years ago

Last modified 5 years ago

#11005 closed bug (duplicate)

WebPositive _ZSt15__once_callable

Reported by: bbjimmy Owned by: pulkomandy
Priority: blocker Milestone: R1/beta1
Component: Applications/WebPositive Version: R1/Development
Keywords: Cc:
Blocked By: #10938 Blocking:
Has a Patch: no Platform: All

Description

On hrev47467

x86 GCC 2 Hybrid

WebPositive will not launch.

Could not open "WebPositive" (Missing Symbol: _ZSt15once_callable)

Attachments (2)

haikuwebkit.log (3.4 KB ) - added by jessicah 5 years ago.
0001-Force-libstdc-to-use-TLS.patch (1021 bytes ) - added by jessicah 5 years ago.

Download all attachments as: .zip

Change History (14)

comment:1 by zooey, 5 years ago

I have checked if this is an artefact of the recent changes to the way the nightly images are being built, but apparently it isn't:

The libWebKit package available for x86_gcc2_hybrid seems to have been built against a Haiku whose libstdc++.so exported the _once_callable symbol, which is no longer the case (since the activation of --enable-tls in the buildtools). A new libWebKit package, built on current Haiku should fix this.

comment:2 by jessicah, 5 years ago

I've rebuilt webkit on a current nightly, hrev47471 to be exact, and still the same unresolved symbol error.

comment:3 by taos, 5 years ago

Strange, I have now a working WebPositive with libwebkit 1.4.1 in hrev47467 built on hrev47449-gcc2-hybrid (see #10938).

comment:4 by jessicah, 5 years ago

I repackaged the gcc package on my Haiku install to use a symlink to the system libstdc++.so, and now haikuwebkit fails to link, expecting the _once_callable symbol to be present. So perhaps we need to change buildtools for cross-building? I shall try this next.

by jessicah, 5 years ago

Attachment: haikuwebkit.log added

by jessicah, 5 years ago

comment:5 by jessicah, 5 years ago

The attached patch is just a draft at this stage. However, after building buildtools and haiku after making this change, WebPositive now launches without errors on hrev47470 x86gcc2_hybrid.

Anyway, as I mentioned on IRC, I definitely believe _once_callable is the correct exported symbol, not _once_functor, given grokking the sources, e.g. http://www.radford.edu/~acm/midatl/docs/C++/a01272.html.

I don't know what's the proper way to enable native TLS support without modifying an OS header like in the patch; hopefully someone else can dig a bit deeper into this particular issue?

comment:6 by waddlesplash, 5 years ago

Milestone: R1R1/alpha5
Priority: normalblocker

comment:7 by jscipione, 5 years ago

Any progress here? This bug is crippling me.

comment:8 by waddlesplash, 5 years ago

My memory is flaky, but I recall Oliver asking Axel on best procedures to migrate our libstdc++ files to their own package, as well as removing some hacks (apparently we convert a static library to a shared one in this process :S).

comment:9 by ttcoder, 5 years ago

Cc: degea@… added

comment:10 by pulkomandy, 5 years ago

Blocked By: 10938 added
Resolution: duplicate
Status: newclosed

comment:11 by ttcoder, 5 years ago

Cc: degea@… removed

comment:12 by pulkomandy, 5 years ago

Milestone: R1/alpha5R1/beta1
Note: See TracTickets for help on using tickets.