|Submitter:||msofer||Created on:||2007-10-13 00:23:11|
|Subsystem:||47. Bytecode Compiler||Assigned To:||msofer|
|Status:||Open||Last Modified:||2008-07-30 06:30:04|
In 8.4 and 8.5: % set x 1; set x [proc set args "return $x"] % puts $x % set x 1 What it should be by rule #2 in Tcl.n: % set x 1; set x [proc set args "return $x"] % puts $x ;# [set] modified before being called 1 % set x 1
msofer added on 2008-07-30 06:30:04:
Logged In: YES user_id=148712 Originator: YES Lowering prio: can't be fixed without a change in Tcl.n (or a huge slowdown), we'll live with this one for a while.
dkf added on 2007-10-14 02:14:53:
Logged In: YES user_id=79902 Originator: NO Quite apart from the fact that your example confuses me anyway, it's an example of something that's deeply horrific anyway. And it's impossible to fix without ridiculous contortions; saying that the command to be invoked is looked up before substitution of the arguments is probably sanest. (On the other hand, that gets horrible when commands are autoloaded, as there the substitution order relative to the autoload step is different, with the compiled command being changed as a side-effect. Ugh.) Since in practice only insane code does this (yes, that makes test suite writers very strange people) I can't see that this is release-critical. It's just one of the Crazy Aunts in Tcl's Attic.
msofer added on 2007-10-13 07:57:32:
Logged In: YES user_id=148712 Originator: YES This is a tough one to fix. One way out is by subtly changing Tcl.n's Rule #2 to ignore the redefinition of a command during the substitution of its arguments, ie Rule 2': A command is evaluated in three steps. 1. the interpreter substitutes the first word, determines the corresponding command 2. the interpreter substitutes the other words 3. the command is invoked