291123 - remove unused MOZ_THREADSAFE_RDF and locks from nsInMemoryDataSource

Status: RESOLVED FIXED

(Core Graveyard :: RDF, defect)

Description

[Axel Hecht]

nsInMemoryDataSource has code fragments that claim to enable multi-threading locks
to the datasource.
They may even work, but they haven't been tested for years, I didn't even try
compiling those.
Anyway, the new API design has interfaces like rdfITripleVisitor,
http://wiki.mozilla.org/Rdf:Interfaces#rdfITripleVisitor
which is called for each arc in the result of a datasource query.
We need to protect the datasource from write operations during this operation.

In a multithreaded env this is tricky->hard, as a write from a read callback is
an error, while a write from a separate thread as a read is just supposed to 
wait.
As we don't use it, I'd like to get rid of it completely, so that we can just
error if we write during a read.

Comment 1

[Axel Hecht]

Attachment: remove unused lock code from nsInMemoryDataSource

Comment 2

[Benjamin Smedberg]

Comment on attachment 181288 [details] [diff] [review]
remove unused lock code from nsInMemoryDataSource

Do you want to remove the "Locked*" function names while you're at it?

Comment 3

[Axel Hecht]

(In reply to comment #2)
> Do you want to remove the "Locked*" function names while you're at it?

I did think about renaming them to something like InnerAssert or something, but
I didn't feel like that is really necessary.
I wouldn't remove them, as they are factored out from the Assert, the Move and the
Change code.

Comment 4

[Brendan Eich [:brendan]]

Comment on attachment 181288 [details] [diff] [review]
remove unused lock code from nsInMemoryDataSource

rs=[e-mail].

/be

Comment 5

[Axel Hecht]

Comment on attachment 181288 [details] [diff] [review]
remove unused lock code from nsInMemoryDataSource

requesting approval, this patch is removing code NPOB.

Comment 6

[Brendan Eich [:brendan]]

Comment on attachment 181288 [details] [diff] [review]
remove unused lock code from nsInMemoryDataSource

Back end code, so 1.8b2+ but I'll throw in aviary1.1a1+ too!

/be

Comment 7

[Darin Fisher]

Drive by nit: The names "LockedAssert" and "LockedUnassert" seem to have been
chosen to reflect the fact that they are called while mLock is held.  Since you
are removing mLock, shouldn't you change the names of these functions?  Maybe
"InternalAssert" or something like that would be better?

Comment 8

[Axel Hecht]

There are more "drive by" nits to fix in this code, I prefer to land them in 
a distinct cycle, when I don't have other patches in my trees.