On Mon, 9 Jul 2012, Mark Morgan Lloyd wrote:
Mark Morgan Lloyd wrote:
Tomas Hajny wrote:
On Wed, July 4, 2012 11:32, Mark Morgan Lloyd wrote:
When building a shared library, I'm putting a (function that returns a)
magic number into both the library and the calling program. This returns
an integer, and allows the program and library to check their exposed
APIs without risk.
How can code in the library test whether it is using cmem, so that it
can tell the caller that it's safe to call functions that move strings
and objects around? It's obviously trivial to rely on a compile-time
conditional, but can this be done in a way that doesn't rely on this?
First of all, you can easily check whether the default memory manager is
used by calling IsMemoryManagerSet. Obviously, this doesn't tell whether
the custom memory manager is cmem or something else (e.g. TraceManager
from HeapTrace). However, you could possibly try to check the distance
between the individual MemoryManager methods retrieved using
GetMemoryManager - while that is not bulletproof by any means, it provides
at least some indication.
Finally, if there are certain major changes to features provided by
individual memory managers (like some operation being safe or not), adding
some TMemoryManager method allowing to provide this information may be the
best solution for the future.
Thanks Tomas (and Ludo). Under the circumstances I think IsMemoryManagerSet
is probably an adequate solution, since the real risk is that somebody
messes about with the start of the main unit (or with the project settings)
without appreciating that the cmem import is crucial to operation elsewhere
in the program (or, analogously, that a shared library will crash when some
functions are called).
Having something- e.g. an exported string- in cmem would be a partial
solution, but would have to be checked in the main unit. With any solution,
I think that testing that cmem was the first import could be a challenge.
I've added a feature request as
http://mantis.freepascal.org/view.php?id=22386 but am being told "First of
all, you are wrong in thinking that using Cmem is a prerequisite to being
able to use longstrings etc. as parameters and results..."
Obviously I accept that there might be other suitable memory managers, but
cmem is the one normally referred to. Far be it from me to argue with those
more experienced than I, but the suggestion I'm trying to make is that it
would be desirable if each module (i.e. every shared library plus the main
program) could confirm that it's in a state where it can share e.g.
longstrings, objects etc. with others. In common usage, that implies that
cmem is not only imported, but is imported /first/.
Ignoring for a second that other possibilities exist:
If the interface of your DLL requires the use of cmem, it is the programmer's
duty to ensure that cmem is used.
I don't even see why you would want to check it, since your interface definition
implies it. In that case, not using cmem is simply not an option, and if the
programmer does not include it properly, it's his fault:
in that case he provided a faulty DLL.
He could equally well be exporting a function with a correct name, but a wrong
set of parameters. You can't check that either and it probably would not occur
to you to do so in the first place...
To me, your cmem unit problem falls in the same category, and consequently
I do not think a solution in the RTL is needed.
Michael.
_______________________________________________
fpc-pascal maillist - fpc-pascal@lists.freepascal.org
http://lists.freepascal.org/mailman/listinfo/fpc-pascal
