Updated 2018-10-11 11:51:47 by stevel

A gotcha is an unexpected side effect, behavior, consequence, requirement, etc.

Description  edit

As with any language, the syntax and semantics of Tcl can catch a programmer off guard. This page is enumerates constructs and behaviours that can be considered as gotchas, and is somewhat arranged by prominence and severity.

Gotchas  edit

The Octal Bug

See Tcl and octal numbers

set and namespaces

Twylite 2012-12-12: Setting a variable in a namespace eval can clobber a global variable. See Dangers of creative writing.
set foo 10
namespace eval bar { set foo 20 ; set bar 30 }
puts $foo ;# puts: 20
puts $bar ;# error: can't read "bar": no such variable

Assuming that a value is a dictionary

In the following example the fact that $dict is not a dictionary does not trigger an error:
set dict hello
dict exists $dict Bob ;# -> 0

Assuming that a value is a list

Another Tcl gotcha is to hand arbitrary strings, read from the user or a file/command, directly into a list operation without first ensuring that the contents is, in actuality, a list.

DKF: The best way to deal with such user input, where users aren't expecting to write a Tcl list in the first place, is to use a sanitizing command to convert the input into a proper list. Examples of sanitization commands can include split, splitx and (my favorite) this:
set elements [regexp -all -inline {\S+} $line]

Assuming that a variable is an array

Using array unset on a variable that isn't array does nothing, and does not triggeer an error:
set array 5
array unset array
set array ;# -> 5

lsearch default matching style is -glob

Forgetting to use -exact when that's what is intended can lead to surprises.

switch and its Arguments

RS One possible gotcha is switch -- always use -- before the switch variable, since if the value of the switch variable starts with -, you'll get a syntax error.

AMG: This isn't required by Tcl 8.5 onward.

Delimiting Options from Arguments

KPV Also, comments within switch, while possible, are tricky.

RS 2010-05-10: A similar gotcha is in the text search subcommand - although the misunderstanding could be avoided by counting non-optional arguments from the end,
set whatever -this
$t search $whatever 1.0

raises an error that -this is an undefined switch. For robustness, use
$t search -- $whatever 1.0

if the slightest possibility exists that $whatever might start with a dash.

PYK 2017-05-19: [scripted lists] makes comments within a switch possible, but it needs to move into the Tcl implementation level to be performant.

Interpretations by expr

RS 2010-02-24: Yet another gotcha we ran into last night: Consider a function
proc f x {
    if {$x == 00000000} {
        puts "$x is NULL"
    }
}

which reported:
0E123456 is NULL

How so? Bug? No -- feature. With the == comparison operator, the operands are tried to match as integers, floats, or then as strings. And $x in this case, though meant as a pointer in hex, could be parsed as float - with the numeric value of 0, which is numerically equivalent to 00000000. The solution of course was to use the eq operator instead.

AMG: Another issue with expr interpreting stuff happens when said stuff was already interpreted by Tcl. This creates performance, security, and correctness problems. Always brace your expr-essions!

Another example:
expr {{} < 0} ;# -> 1
expr {{} > 0} ;# -> 0
expr {{} == 0} ;# -> 0

For the < and > operators, {} is interpreted as an integer, 0. For the == operator, 0 is interpreted as a string.

because

string is and the empty string

string is always returns 1 for the empty string. To avoid this, use -strict

Confirming that a value is an integer decimal notation

The string is commands accept all the numeric notations that expr does (see The Octabug, so they aren't up to the job by themselves. Here is one way to do it:
set mynumber 0x11
expr {[string is entier -strict $mynumber] && [scan $mynumber %d mynumber] > 0}

Partial sub-command resolution in namespace ensembles

namespace eval table {
    namespace export *
    namespace ensemble create
}

proc {table::spoon fork} x {
    return $x
}

Even though there is no spoon, there is:
% puts [table spoon huh?]
huh?

To change this behavior:
namespace ensemble configure table -prefixes no

Nested vwait  edit

See Avoiding Conflicting vwait's on the vwait page

Starving Queues with after  edit

See after x after idle on the after page

See Also  edit

Dangerous constructs

Larry Smith I was expecting a new control construct of some sort.
 gotcha {
   ...
 } now {...}