Re: Error reporting should be improved

[Bob Friesenhahn]

On Wed, 30 Dec 2009, Peter Rosin wrote:
> Aha, you have misunderstood the patch, that's why we don't understand
> each other. The patch does indeed allocate a buffer, that's the whole
> point of sending the error message to lt_dladderror. The error message
> is cached indefinitely inside lt_dladderror. In order to not grow that

In what line of code is a string buffer allocated?  The only buffer 
allocation I see is a REALLOC to increase the buffer which holds 
diagnostic pointers.  Then the passed pointer is assigned directly to 
the pointer at errindex without any allocation or string copy.

> "cache", I implemented a comparison so that registering the same user
> error multiple times results in returning the index of the old error
> message instead of inventing a new error number each time.

Yes.  And so it is impossible to decipher meaningless errors from 
meaningful ones.  If the error comes from a loader which should not 
have worked in the first place, then the user should be able to ignore 
it.

> Note that user_error_strings is an array of strings, not one big long
> string.

Yes, it is an array of string pointers.

> The patch does exactly the first part, but not the second. There is simply
> no way to know when the user is done with the string, so it has to be
> kept indefinitely. We could document the lifetime rules, but until then
> there is no way to know when the message can be freed.

The documentation should say that this data is only valid until the 
next lt_* call, which is currently the case.

> Yes, that is true. But that is not a new problem, just magnified by
> increased usage of lt_dladderror.

It was essentially unused before since it was only used by a rarely 
used loader on OS-X.

This discussion should completely move to the patches list so I will 
delete the Cc: to the bugs list.

Bob
--
Bob Friesenhahn
address@hidden, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/