H5Pget_driver_info: do not push an error when the driver has no info block

[hyoklee]

Fixes #6651.

H5Pget_driver_info() pushed H5E_CANTGET whenever H5P_peek_driver_info() returned NULL. NULL is not necessarily a failure there: a driver that registered no driver-specific properties leaves driver_info NULL, which is exactly what H5Pset_driver_by_name() and the HDF5_DRIVER environment variable produce.

Both the function's own header comment and its entry in H5Ppublic.h say that case returns NULL with no error pushed:

Failure: NULL. Null is also returned if the driver has not registered any driver-specific properties although no error is pushed on the stack in this case.

and

If no driver-specific properties have been registered, H5Pget_driver_info() returns NULL.

The sibling H5Pget_driver_config_str() already handles the same situation correctly:

    if ((config_str = H5P_peek_driver_config_str(plist))) {

Effect

Any VFD that fetches its optional FAPL info block — including HDF5's own H5FDmulti.c, in four places — prints an HDF5-DIAG stack on every open describing an error that did not occur:

HDF5-DIAG: Error detected in HDF5 (2.3.0):
  #000: src/H5Pfapl.c line 1546 in H5Pget_driver_info(): can't get driver info
    major: Property lists
    minor: Can't get value

I hit this running the netCDF-C test suite over an out-of-tree VFD selected with HDF5_DRIVER=<name>: every H5Fopen in the run produced one of these. Harmless to the operation, but not to testing — any test comparing a program's stderr against a reference sees a diff, and someone reading the output of a driver under development is told there is an error when there is not.

The change

Drop the wrapper HGOTO_ERROR and assign the result. Nothing that was diagnosed stops being diagnosed: H5P_peek_driver_info() pushes H5E_BADTYPE itself for a list that is not a file access property list, and H5E_CANTGET when the property cannot be read. The removed frame only ever duplicated one of those or fired on a normal NULL.

Not a recent regression — the code dates to at least [svn-r15510] (2008).

🤖 Generated with Claude Code

[github-actions]

Review Checklist

This PR touches the following areas. Each needs a sign-off
from its listed owners before merging.

• src
  • @fortnern (manually added) — approval required

[hyoklee]

Verified with a minimal reproducer. H5Pset_driver_by_name(fapl, "sec2", NULL) is what HDF5_DRIVER=sec2 does internally and sets no driver-info block, so driver_info is legitimately NULL:

if ((fapl = H5Pcreate(H5P_FILE_ACCESS)) < 0) return 1;
if (H5Pset_driver_by_name(fapl, "sec2", NULL) < 0) return 1;

info  = H5Pget_driver_info(fapl);
nerrs = H5Eget_num(H5E_DEFAULT);

Before, on develop @ b7b85e7abf9, Windows 11 / MSVC 19.x:

HDF5-DIAG: Error detected in HDF5 (2.3.0):
  #000: src\H5Pfapl.c line 1546 in H5Pget_driver_info(): can't get driver info
    major: Property lists
    minor: Can't get value
driver_info = NULL (no info block)
errors on the stack after the call = 1

After:

driver_info = NULL (no info block)
errors on the stack after the call = 0

Same return value, and the stack is now clean — which is what the documented contract says should have been happening all along.