[
https://issues.apache.org/jira/browse/MRESOLVER-179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17339083#comment-17339083
]
Guillaume Nodet commented on MRESOLVER-179:
-------------------------------------------
I tend to think the new API is too fine grained. In addition to the above
problem, in my use case, having multiple calls to {{NamedLock}} would lead to
multiple inter-process calls, while a single call could be sufficient. I
wonder if the NamedLock should be kept as an internal implementation detail of
the {{SyncContextFactory}}.
> NamedLocks should be acquired in an ordered way
> -----------------------------------------------
>
> Key: MRESOLVER-179
> URL: https://issues.apache.org/jira/browse/MRESOLVER-179
> Project: Maven Resolver
> Issue Type: Bug
> Reporter: Guillaume Nodet
> Priority: Major
>
> [~michael-o] [~cstamas] I was about to start porting my simple SyncContext to
> maven-resolver and start working on a PR when I saw that the API has changed
> a lot with the introduction of NamedLocks.
> I see some design problems in the new API. Multiple locks are now acquired in
> sequence instead of using a coarse grained call. This can lead to deadlocks
> as one client could try to lock {{a}} then {{b}} while another one could try
> to lock {{b}} then {{a}}, leading to a possible deadlock.
> It should be just a matter to order them inside the
> {{AdaptedLockSyncContext.acquire}} call which is the only location where the
> API is used.
> I'll provide a PR asap.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
