* ABI's to break:
** gpgme_data_new_from_filepart takes an off_t as count, but should
   take a size_t.
** GpgmePassphraseCb should have void **R_HD, not void *R_HD.
** trustlist has the same start/end problem as keylist had.
   In fact, all the _start functions have this problem!
   In addition, the resulting error of the operation can not be
   retrieved seperately; the op_foobar operations can't be implemented
   by the user, they are not merely convenience, but necessity, while
   the op_foobar_start functions for these are unusable (or render the
   context unusable, your choice).
** string representation of non-secret keys and ATTR_IS_SECRET is NULL,
   which can not be differentiated from the case that it is not
   representable.
** Agree on gpgme_key_unref or gpgme_key_release and drop the other?

* Implement posix-sema.c

* Allow to use GTK's main loop instead of the select stuff in
  wait.c

* add locking to the key cache?

* Should --delete silently delete secret keys or is there a need for
  another flag or a callback?

* GpgmeKey misses GPGME_ATTR_EXPIRE attribute

* Add ATTR to return the number of subkeys or uids.

* Return GPGME_Canceled when appropriate

* Factor out common code in _op_*_start functions.

* Documentation
** Add note about GPGME clearing out pointer return values.
** validity/trust

* Engines
** Move code common to all engines up from gpg to engine.
** engine operations can return General Error on unknown protocol
   (it's an internal error, as select_protocol checks already).

* Error Values
** Map ASSUAN error values.
** Map GpgSM ERR messages.
** Verify (and document) if Read_Error, Write_Error, Pipe_Error set errno.

* Tests
** t-data
*** Test gpgme_data_release_and_get_mem.
*** Test gpgme_data_rewind for invalid types.
*** Test gpgme_data_read's readable feature.

* Build suite
** Make sure everything is cleaned correctly (esp. test area).

Bugs reported by Stephane Corthesy:
> - When returning a GpgmeKey GPGME_ATTR_COMMENT attribute, characters  
> like ":" are not un-escaped, they are returned as \x3a

> BTW, here's another bug: it it not possible to retrieve fingerprints  
> for subkeys

> In GpgmeRecipients, would it be possible to provide a function which  
> would return the validity assigned to a name contained in the  
> GpgmeRecipients instance?

> - There is an inconsistent behaviour: if we pass three times an  
> invalid (but non empty) passphrase, return code is GPGME_No_Data, but  
> if we pass three times an empty (and invalid) passphrase, we get  
> GPGME_No_Passphrase.

> passphrase callback. If I use the same GpgmeContext as the one which  
> is currently asking for a passphrase, my app crashes: the r_hd in
> the  
> callback has become invalid; if I use a brand new one, the callback  
> is called recursively, when I ask to enumerate keys.

> Talking about gpgme performances: did anyone make some profiling on
> gpgme calls and can tell me why it takes so long to enumerate the
> whole pubring? Listing keys with gpg is very fast, whereas with
> gpgme_op_keylist_XXX() it's soooooo slow.