These changes address two related problems with baton, firstly
inconsistent memory management on errors, leading to segfaults and
secondly, memory management code duplication between error and
non-error code paths.
init_rods_path (called from resolve_rods_path) is no longer
responsible for memset'ing the rods_path struct. Now the caller does
this immediately after declaring. This in repetitive, but is simple
and eliminates the most common class of bug found recently in
baton (segfaults due to uninitialised memory on the error handling
code path).
Where non-error and error code paths share operations and return
values, these have been combined with a new label 'finally'.
As the changes are extensive, the API has been tested with
Valgrind. Included is a new wrapper script for baton-do that can be
used where higher level APIs run a baton-do child process. This allows
the specific baton-do functions used by the higher level API to be run
under Valgrind.
The Valgrind tests revealed an existing memory leak in list_checksum,
which has been fixed.
These changes address two related problems with baton, firstly inconsistent memory management on errors, leading to segfaults and secondly, memory management code duplication between error and non-error code paths.
init_rods_path(called fromresolve_rods_path) is no longer responsible for memset'ing therods_pathstruct. Now the caller does this immediately after declaring. This in repetitive, but is simple and eliminates the most common class of bug found recently in baton (segfaults due to uninitialised memory on the error handling code path).Where non-error and error code paths share operations and return values, these have been combined with a new label 'finally'.
As the changes are extensive, the API has been tested with Valgrind. Included is a new wrapper script for baton-do that can be used where higher level APIs run a baton-do child process. This allows the specific baton-do functions used by the higher level API to be run under Valgrind.
The Valgrind tests revealed an existing memory leak in list_checksum, which has been fixed.