Tcl Source Code

View Ticket
Bounty program for improvements to Tcl and certain Tcl packages.
Tcl 2019 Conference, Houston/TX, US, Nov 4-8
Send your abstracts to [email protected]
or submit via the online form by Sep 9.
Ticket UUID: 4f322b9d2112772a4a9599611def6e651f8e5198
Title: Impossibility to create an unique path in Tcl (e. g. using file mkdir)
Type: RFE Version: >= 8.5
Submitter: sebres Created on: 2018-07-02 12:12:33
Subsystem: 16. Commands A-H Assigned To: nobody
Priority: 5 Medium Severity: Minor
Status: Pending Last Modified: 2018-10-10 22:41:54
Resolution: Fixed Closed By: nobody
    Closed on:

If working multi-threaded (or multi-processed in cluster-system on share) with some global directory to create some system-wide unique sub-path using some pseudo-unique name generation algorithm (also UUID), to avoid retrieving of the same folder for two workers, one can do the things like this:

proc get_unique_dir {} {
  while 1 {
    set p [file join $GLOB_PATH [some_uuid_algo]]
    if {![file exists $p]} {
      file mkdir $p
  return $p

But it's affected by the race condition (firstly not exists for both workers, then one creates sub-path, another returns with no error in `file mkdir` on same existing path), if some_uuid_algo cannot guarantee a system-wide unique path.

So the creation of a safe construct for that is impossible in Tcl currently.


Currently `file mkdir` returns no result at all.
If `TclFileMakeDirsCmd` will be extended with return value as counter (of created tail-directories), one may use it to avoid race condition in above-mentioned script, like here:

proc get_unique_dir {} {
  while 1 {
    set p [file join $GLOB_PATH [some_uuid_algo]]
    if {[file mkdir $p]} {
  return $p

This is race condition safe. The only one backwards-incompatibility in this case would be usage like this:

set some_var "when [file mkdir $some_path]was used in substitution"
But I believe the way like this is very rarely (or almost never) used.

User Comments: sebres added on 2018-10-10 22:41:54:

So [270f78ca95b642fb] is definitely closed, well it is time to "close" another race-condition (this RFE fixed).
Someone with an idea to bring this thing further?
I spoke to several people: no one has used [file mkdir] to get/check empty return value / or somehow expected empty values in string like "...[file mkdir]..." or similar imaginable substitutions.
Any suggetions, proposals, etc?

aspect added on 2018-07-12 09:43:10:
I approve of this change.  Can we afford to break code that relies on the
empty result from [file mkdir]?

See also and the ensuing
wiki discussion at is also relevant.

sebres added on 2018-07-02 15:05:03:

Implemented in branch rfe-4f322b9d21 for >= 8.5th.

Although I think, it could be accepted as it's, adding a short changelog entry with small incompatibility note like: command `file mkdir` returns integer now vs. nothing previously... But TCT should decide about it.

So pending as implemented.