| 00:54.46 | *** join/#stepcode mpictor (~mark@c-68-58-38-45.hsd1.in.comcast.net) |
| 00:56.32 | *** join/#stepcode hoiji (671b082c@gateway/web/cgi-irc/kiwiirc.com/ip.103.27.8.44) |
| 01:14.14 | mpictor | hi hoiji |
| 01:14.23 | hoiji | hi |
| 01:14.32 | mpictor | how are you doing? |
| 01:16.01 | hoiji | can you help me with this http://infobot.rikers.org/%23stepcode/20140723.html.gz |
| 01:16.16 | hoiji | I think I asked this when u were offline |
| 01:16.39 | *** part/#stepcode hoiji (671b082c@gateway/web/cgi-irc/kiwiirc.com/ip.103.27.8.44) |
| 01:16.52 | *** join/#stepcode hoiji (671b082c@gateway/web/cgi-irc/kiwiirc.com/ip.103.27.8.44) |
| 01:23.29 | hoiji | As for my status: in clstepcore library is complete apart from classes EntList (+ its subclasses) , STEPcomplex |
| 01:25.52 | hoiji | The problem (with STEPcomplex class) is quite similar to EntList class. |
| 01:30.01 | hoiji | My solution in a nutshell is a variant of hand-to-hand locking |
| 01:30.55 | hoiji | A typical hand-to-hand locking is of the following form |
| 01:35.26 | hoiji | cur = head; cur->mtx.lock(); |
| 01:35.38 | hoiji | while( <some condition on cur> ) { |
| 01:35.45 | hoiji | <Do some work on cur>; |
| 01:36.00 | hoiji | cur = cur->next; |
| 01:36.24 | hoiji | cur->mtx.lock(); // see at this point we have two lock |
| 01:36.31 | hoiji | cur->prev->unlock(); |
| 01:36.32 | hoiji | } |
| 01:37.44 | hoiji | The problem with the above approach is that it could lead to deadlock if there are concurrent forward and backward iterations |
| 01:38.15 | hoiji | So I have devised a new approach (basically a variant) |
| 01:40.40 | hoiji | cur = head; |
| 01:40.54 | hoiji | while( <some condition on cur> ) { |
| 01:41.02 | hoiji | cur->mtx.lock(); |
| 01:41.13 | hoiji | <Do some work on cur>; |
| 01:41.21 | hoiji | cur = cur->next; |
| 01:41.26 | hoiji | cur->prev->mtx.unlock(); // see at this point we have No locks |
| 01:41.42 | hoiji | cur->mtx.lock(); |
| 01:41.47 | hoiji | } |
| 01:42.51 | hoiji | The above variant will fail badly if there the List API allows for removing a node or inserting a node (In between two nodes) |
| 01:45.05 | hoiji | Since currently in both STEPcomplex and EntList data structures there are only search and append operations (i.e insert after last node), the above variation will not lead to any problems |
| 01:48.17 | hoiji | In the above approach there is a window where a thread will be holding no locks, hence deadlocks wont happen in forward / backward iterations |
| 03:08.01 | hoiji | note: the hand-to-hand locking I am referring to is more popularly knows as hand-over-hand |
| 03:16.25 | mpictor | I meant to check back in a minute but got distracted :( |
| 03:17.38 | mpictor | reads the log |
| 03:18.54 | hoiji | :) |
| 03:20.53 | mpictor | that sounds ok to me, just add a comment to the class noting the issue with adding those functions |
| 03:21.06 | mpictor | *methods |
| 03:23.00 | hoiji | thanks |
| 03:24.05 | mpictor | any time :) |
| 03:26.47 | mpictor | I will check in the morning to see if you have posted anything else |
| 03:27.18 | mpictor | my morning, that is |
| 03:41.10 | hoiji | haha, :D On a side note my semester has started, am going for my class now, The classes today will last probably till your morning :P |
| 08:09.17 | *** join/#stepcode hoiji (671b082c@gateway/web/cgi-irc/kiwiirc.com/ip.103.27.8.44) |
| 17:47.22 | *** join/#stepcode github-notify (~github-no@192.30.252.35) |
| 17:47.22 | github-notify | [stepcode] hoiji09 pushed 3 new commits to hj/thread-safety: https://github.com/stepcode/stepcode/compare/06b7ee582363...3fdbe460f14a |
| 17:47.22 | github-notify | stepcode/hj/thread-safety 7233d60 hoiji09: Fixed locking in sdaiApplication_instance.cc... |
| 17:47.22 | github-notify | stepcode/hj/thread-safety 2cabefa hoiji09: Minor STEPcomplex optimization... |
| 17:47.22 | github-notify | stepcode/hj/thread-safety 3fdbe46 hoiji09: Made STEPcomplex thread safe... |
| 17:47.22 | *** part/#stepcode github-notify (~github-no@192.30.252.35) |