Revision history [back]

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

Failure seems to found!!! Building ros_cpp with "-Wcast-align" shows the possible lines... it seems that casts from uint8 to uint32 causes the alignment errors!! Ok... failure seems to be figured out, but how do i solve this???

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

Failure seems to found!!! Building ros_cpp with "-Wcast-align" shows the possible lines... it seems that casts from uint8 to uint32 causes the alignment errors!! Ok... failure seems to be figured out, but how do i solve this???

Ok... rebuilding everything with flag "-mno-unaligned-access" ... i hope it works this time and is a more correct solution then putting everything to debug... i will let you know if it is stable this time...

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

Failure seems to found!!! Building ros_cpp with "-Wcast-align" shows the possible lines... it seems that casts from uint8 to uint32 causes the alignment errors!! Ok... failure seems to be figured out, but how do i solve this???

Ok... rebuilding everything with flag "-mno-unaligned-access" ... i hope it works this time and is a more correct solution then putting everything to debug... i will let you know if it is stable this time...

as info: since arm-v4 arm arch is able to handle unaligned access. cortex-a8/9 are arm-v7a, so they can handle, except there is a NEON are or something else... similiar... if there is a NEON compiler make use of faster memory access and so on. and there is the problem: NEON only handles ALIGNEND memory addresses. thats what i figured out, if i'm wrong please correct me!!

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

Failure seems to found!!! Building ros_cpp with "-Wcast-align" shows the possible lines... it seems that casts from uint8 to uint32 causes the alignment errors!! Ok... failure seems to be figured out, but how do i solve this???

Ok... rebuilding everything with flag "-mno-unaligned-access" ... i hope it works this time and is a more correct solution then putting everything to debug... i will let you know if it is stable this time...

as info: since arm-v4 arm arch is able to handle unaligned access. cortex-a8/9 are arm-v7a, so they can handle, except there is a NEON or something similiar... if there is a NEON compiler make use of faster memory access and so on. and there is the problem: NEON only handles ALIGNEND memory addresses. thats what i figured out, if i'm wrong please correct me!!

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

Failure seems to found!!! Building ros_cpp with "-Wcast-align" shows the possible lines... it seems that casts from uint8 to uint32 causes the alignment errors!! Ok... failure seems to be figured out, but how do i solve this???

Ok... rebuilding everything with flag "-mno-unaligned-access" ... i hope it works this time and is a more correct solution then putting everything to debug... i will let you know if it is stable this time...

as info: since arm-v4 arm arch is able to handle unaligned access. cortex-a8/9 are arm-v7a, so they can handle, except there is a NEON or something similiar... if there is a NEON compiler make use of faster memory access and so on. and there is the problem: NEON only handles ALIGNEND memory addresses. thats what i figured out, if i'm wrong please correct me!!

:'( ... even rebuilding stacks with no unaligned access didn't work... thats disappointing... can plz someone report if all its build nodes work proper!? especially nodes high-level nodes like gmapping!

I also build everything from robot package including pcl (but except openni) on armhf (pandaboard)... compiling was not the problem at all. Build it several times, and always got failures in different packages while running them... always misalignment during deserialization. it occurs when node subscribers are connected and the first message is deserialized. Now i would like to know if this problem also occurs on raspberryPI?

this alignment error makes armhf on pandaboard UNUSABLE!! :-( currently i'm downgrading to 12.04 armel losing all advantes of hard-float support, but when i tried everything with ubuntu 11.10 armel everything worked fine for me... so hopefully i have a working panda with an ubuntu 12.04 this evening...

some users in forum said putting build to debug solves the problem but this wasn't correct for me. i think it's just luck if compiler addresses variables in program correctly so that they would fit to armhf support... i think there is the problem. or are there any other suggestion? or may be a workaround?

Next day, next try... found some infos to alignment error will rebuild evrything on armhf .... again .... this time building every package itself, activating gcc cast and alignment warnings.... saw that some people have same problem, not only related to ros. Some others figured out that aligment error depends on gcc usage of fast memory operations resulting in unhandled kernel exceptions... hope that i can figure out the problem... will start with serialization... ros_comm and bond...

Failure seems to found!!! Building ros_cpp with "-Wcast-align" shows the possible lines... it seems that casts from uint8 to uint32 causes the alignment errors!! Ok... failure seems to be figured out, but how do i solve this???

Ok... rebuilding everything with flag "-mno-unaligned-access" ... i hope it works this time and is a more correct solution then putting everything to debug... i will let you know if it is stable this time...

as info: since arm-v4 arm arch is able to handle unaligned access. cortex-a8/9 are arm-v7a, so they can handle, except there is a NEON or something similiar... if there is a NEON compiler make use of faster memory access and so on. and there is the problem: NEON only handles ALIGNEND memory addresses. thats what i figured out, if i'm wrong please correct me!!

:'( ... even rebuilding stacks with no unaligned access didn't work... thats disappointing... can plz someone report if all its build nodes work proper!? especially nodes high-level nodes like gmapping!

back again... and happy now! didn't found a solution fitting to all probs, but possibly it may help: dont use serialization! try to build nodelets and load them into a manager! manager are multi-threaded at all and can also distribute multi-threaded node-handles! due to intra-process-communication no de-/serialization is needed -> no alignment fault should happen!