| Ticket UUID: | 62144fdc02ef37b0fcbdbbcd05ffb8375446f536 | ||
| Title: | "scan NaN %f" returns a list with an empty element | ||
| Type: | Bug | Created on: | 2026-07-21 11:00:57 |
| Submitter: | oehhar | Assigned to: | nobody |
| Subsystem: | 18. Commands M-Z | Severity: | Minor |
| Priority: | 5 Medium | Last modified: | 2026-07-24 09:21:25 |
| Status: | Open | Closed by: | nobody |
| Resolution: | None | Closed on: | |
| Version: | 8.6-9.1 | ||
| Description: | ||||
|
I tried a bit with doubles, scan and format with NaN and stumbled over the following:
% scan NaN %f
{}
% scan NaN %f fl
0
% set fl
can't read "fl": no such variable
For sure it is a bug:
As NaN is a valid value for a double, it would be beneficial for me, that "scan NaN %f" really scans a double with a NaN value. Apparently, TCL does not see "NaN" as a double value. "INF" works for scan, format, expr. Even if there is no way to generate a NaN within TCL, it may come from an external source like a data base. Thanks for all, Harald | ||||
| User Comments: | ||||
kuehro added on 2026-07-24 07:45:26:
There are two ways to fix the deficient math of our beloved Tcl:
1) Build your own Tcl with line 26 of generic/tclInt.h changed to
#define ACCEPT_NAN
2) Download mymath.tcl from
https://gist.github.com/yyamasak/b1ce55c40631b1c1ea6e8200dba3f7e0
and change line 109 from
if {$res eq "domain error: argument not in valid range"} {
to
if {$res eq "domain error: argument not in valid range" || \
[string first "non-numeric" $res] > 0} {
Now you can use myexpr instead of expr, for example with 8.6.18 and 9.1b1 we get:
% set $n -nan
-nan
% puts [myexpr {2.0+$n}]
NaN
% puts [myexpr {1.0/0.0}]
Inf
% puts [myexpr {0.0/0.0}]
NaN
Searching for nan at https://core.tcl-lang.org/tips/doc/trunk/index.md
shows several TIPs related to proper math handling (520,521, 731, ...).
Method 1) still has inconsequent handling of the sign, i.e. -NaN vs NaN
and method 2) can be used without recompiling but needs myexpr instead
of expr.
Working in a scientific workplace I had to abandon the use of Tcl for this
reason but for many use cases method 1) or 2) might be viable.
Good luck,
Kai-Uwe
oehhar added on 2026-07-24 07:58:45:
Thanks, Kai-Uwe, for the great hints and TIP pointers. I have apparently hit a difficult point. That explains why nobody adds comments. But the reason of this ticket is the return of an list with an emty string element by scan NaN %f So, NaN is represented here by the empty string. But scan NaN %f fl does not scan NaN as the empty string. This is inconsistent. Sorry to hear, that you had to abandon TCL due to missing nan support. Yes, there is now a "isnan()" command which might be the first step. And the documentation of scan/format do not mention +/-Inf neither, which works. Thanks for all, Harald kuehro added on 2026-07-24 08:56:42:
Using a fossil checkout from today and compilation with #define ACCEPT_NAN leads to:
(tcltest) 1 % scan NaN %f
NaN
(tcltest) 2 % scan NaN %f fl
1
(tcltest) 3 % set fl
NaN
(tcltest) 4 % scan -nan %f fl
-NaN
(tcltest) 15 % expr {isnan($fl)}
1
(tcltest) 6 % info patchlevel
9.1b1
Looks Good.
Kai-Uwe
oehhar added on 2026-07-24 09:21:25:
I added a forum to TIP 520 and wrote a message on tcl-core: https://core.tcl-lang.org/tips/forumpost/990033f3 Please use the forum to post any information outside this bug. We risk to loose the information if the bug is closed. Thanks for all, Harald | ||||
