Tcl Source Code

encodings are freed too easily.

Once an encoding has been used
by a program at all, it should be kept
loaded, since it's likely to be needed
again.

As is, the [encoding convertfrom]
command will load an encoding
from a disk file, +count, use it,
-count, and destroy it, all before
returning.  Makes for many too many
disk reads.

Changing the refcounts
should not be too hard, but
there's also work to do updating
the tests in the test suite that
actually insist on this rapid
purge of encoding data.

Logged In: YES 
user_id=80530


heh.  oh well.  Committing for 8.4.12 anyway.

Closing;  will open Tcl_GetEncodingFromObj()
RFE separately.

Logged In: YES 
user_id=80530

this slipped 8.4.10;

raising priority so it makes 8.4.11

File Added - 132026: 1077262-84.patch

Logged In: YES 
user_id=80530


Here's the corresponding patch
for the 8.4 branch.

Logged In: YES 
user_id=80530

patch committed.

Leaving report open as a prompt
to consider moving the new
TclGetEncodingFromObj to
the public interface.

File Added - 129104: 1077262.patch

Logged In: YES 
user_id=80530


attached patch uses a Tcl_ObjType
to cache Tcl_Encoding as an
internal rep, and updates the
[encoding convert*] commands to
use it, fixing the worst of these
problems.

Logged In: YES 
user_id=72656

Perhaps a cache (quasi like REs have) of the last 10 used
encodings get an extra refcount?  That way we have a pointer
to free on exit, and we don't keep all used encodings loaded.