From: Ion Gaztañaga (ion_g_m_at_[hidden])
Date: 2004-11-24 16:51:07
Thank you for the feedback, I have some things to comment:
> Maybe the best way to think about it is how it could be used.
> I see few common scenarios:
> 1. As message queue between producer(s) and consumer(s),
> carying typed or untyped messages (with possible filtering
> of these mesages).
A process shared condition variable would be very useful for this to block
processes when queue is empty/full, do you know any windows implementation
for process-shared condition variables? I think a message queue is a must
for boost. I would love also non-connection oriented message oriented ipc,
something similar to datagram unix socket or a connection oriented stream
ipc, like stream unix sockets. As far as I know cygwin a similar uses TCP/IP
to emulate unix sockets, so I have not seen a pure windows source code
I'm working on the calling syntax, I'm thinking in something like:
segment.named_new<MyType>("name") (Param1, Param2, Param3);
segment.named_new<MyType>("name") [array_length](Param1, Param2, Param3);
to find or create
segment.find_or_named_new<MyType>("name") (Param1, Param2, Param3);
segment.find_or_named_new<MyType>("name") [array_length](Param1, Param2,
No throwing versions:
segment.named_new<MyType>("name", std::nothrow) (Param1, Param2, Param3);
segment.named_new<MyType>("name", std::nothrow) [array_length](Param1,
segment.find_or_named_new<MyType>("name", std::nothrow) (Param1, Param2,
[array_length](Param1, Param2, Param3);
I would like the deletion to be unique although in plan memory new
requires delete, :
segment.delete("MyName"); //does not throw, returns true/false
Do you find them appropiate?
Regarding exceptions, I agree that the return value and the exception are
not a good way to handle the error. If you were using throwing version,
several exceptions can occur:
-> bad_alloc when there is no more memory
-> any exception thrown by any constructor
When using create only version, if the object is already created, I suppose
you would like another exception to be thrown, something like
boost::shmem::already_created. The no throwing versions would return 0 if
any error is found. Is that right?
Regarding find_named_new function I would like to mantain the std::pair<T*,
size_t> version.(not throwing, just true if found).
>>>8. offset_ptr.hpp: full_offset_ptr class
>>>b) The flag could be eliminated completely.
> Hmm, maybe something as (ptrdiff_t)-1 or if there's some symbolic
> name for such value.
A char pointer could point to any byte in the segment, so (ptrdiff_t -1)
would be valid. I can add a new base class for offset_ptr that considers
(ptrdiff_t-1) zero, so that the user can use that pointer if he wants.
>>>12. The simple algorithm to find fitting memory
> Maybe if the library has interface to plug in different
> algorithm (no clue now how it would look like).
I will add another algorith to show that algorithm can be chosen via
>>>13. Can be be possible to identify named objects
>>> with something else than C string? Say
>>> wchar_t* or numeral or other templated type?
>> Do you think that the key type should be templatized?
> Well, this is Boost ;-)
Anyway, the user can create a map with the <key_type, mapped_type> using
actual interface, place it with a name, and use the new map (that can be a
hashed or a tree map) to obtain the same efect.
> This reminds me, newest STLport beta has lock-free allocator
> inside (I know just this, no more details).
I will try to have a look (although it can be added later if the allocation
algorithm is templatized)
> Maybe the library could list those offending STLs and instruction how
> to fix them manually to be useable with shmem. Such a fix should
> not change behaviour of applications, right?
I will add changes needed for Dinkum STL, but I don't know if publishing
template lines and the changes needed can lead to any legal problem.
>>>d) support for "transactions": I would like to
> - transactions with isolation level 1:
> - transactions with isolation level 2:
Sounds like it can be a layer above shmem library, another library perhaps.
I 'll wait some days to see if there are more comments on this library.
After that, I'll try to rewrite some things pointed, make another allocation
algorithm and change the named_new syntax. I will rewrite the documentation
and post another version as soon as possible.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk