Tcl_FSNewNativePath() makes a call to the
Tcl_Filesystem's Tcl_FSInternalToNormalizedProc.
It demands of that driver routine the creation of
a pathObj with string rep (literal or to be created
on demand) of a normalized path name. The
driver doesn't have to do anything more than that.
However, if the driver does do more than that,
caching in the intrep whatever that filesystem
finds useful to keep about its path values,
Tcl_FSNewNativePath() assertively throws all
that away, and constructs a completely new
internal rep for the pathObj.
This doesn't harm the overall function. The
normalized name is preserved, and at least
the claiming filesystem cache is maintained too,
so future operations on the path value will be
correct. However, it does indicate that for some
drivers work will be done only to be undone,
and that's a strange thing for a routine which
is documented as "efficient" to be doing.
I think it would be an improvement if
T_FSNNP() only did its own intrep construction
if the value is not already of the "path" Tcl_ObjType.
If it finds the driver has already set the value to
the "path" Tcl_ObjType, I think the filesystem core
ought to trust the driver to have set its own intrep
as it best sees fit.
|