Side Effects

goodgulf has asked for the wisdom of the Perl Monks concerning the following question:

Replies are listed 'Best First'.
Re: Side Effects by Tanktalus (Canon) on Apr 07, 2005 at 00:00 UTC
Most likely, somehow, $g::TaskList[0] is an alias to $_ (although I'm not entirely sure how that happened). And `while(<LOG>)` does not localise $_, meaning that you've just clobbered $_ through the while. Try putting your body into: `{ local $_; $Body .= "$_<br>" while (<LOG>); }` [download] I'm betting the problem goes away here.	[reply] [d/l] [select]
Re^2: Side Effects by ysth (Canon) on Apr 07, 2005 at 04:21 UTC
The preferred way to do this is `local *_;` (which of course also affects @_, &_, etc.). `local $_;` can have problems if what $_ is aliased to is magic. In 5.10, the preferred way will be `my $_;`.	[reply] [d/l] [select]
Re: Side Effects by ikegami (Patriarch) on Apr 07, 2005 at 04:18 UTC
Unlike foreach loops, `while` does not localize `$_`. In your code, `$_` must be an alias to `$g::TaskList[0]`, like `$array[0]` is in the following code: `@array = ('a'); foreach (@array) { # $_ is an alias for $array[0]. while (<DATA>) { } # $_ and $array[0] are clobbered print("[$_]$/"); # Prints "[]" } print("[$array[0]]$/"); # Prints "[]"` [download] The fix is to localize `$_`: `@array = ('a'); foreach (@array) { # $_ is an alias for $array[0]. local $_ = $_; # This $_ is not an alias for $array[0]. while (<DATA>) { } # $_ is clobbered, but not $array[0] print("[$_]$/"); # Prints "[]" } print("[$array[0]]$/"); # Prints "[a]"` [download] If you want to preserve `$_` within the loop, use: `@array = ('a'); foreach (@array) { # $_ is an alias for $array[0]. { local $_; # This $_ is not an alias for $array[0]. while (<DATA>) { } } # Neither $_ nor $array[0] are clobbered print("[$_]$/"); # Prints "[a]" } print("[$array[0]]$/"); # Prints "[a]"` [download] or: `@array = ('a'); foreach (@array) { # $_ is an alias for $array[0]; while (my $line = <DATA>) { } # Neither $_ nor $array[0] are clobbered print("[$_]$/"); # Prints "[a]" } print("[$array[0]]$/"); # Prints "[a]"` [download]	[reply] [d/l] [select]
Re^2: Side Effects by goodgulf (Initiate) on Apr 07, 2005 at 19:50 UTC
Actually, $g::TaskList is not used, named, or referred to in the entire module that the above code appears in (though it is global). (It is used by an upstream sub that calls down into this subroutine though, but wouldn't $_ at least be local to the subroutine?) Even if $_ somehow became an alias to a global variable that was not used in the module in question, how would using $_ as an r-value undef said global? `$Body .= $_;` [download] How does that undef $_, or anything it might be aliasing to? Sorry if I'm not getting it (I understand the idea that making it local will fix it, but I'd like to understand what makes this not work in the first place, because I just cannot figure out how $_ became an alias to TaskList). Thanks for the continued attention. :)	[reply] [d/l]
Re^3: Side Effects by ikegami (Patriarch) on Apr 07, 2005 at 19:57 UTC
It doesn't, it's the `while(<LOG>)` that clears it. `while(<LOG>)` is short for `while ($_ = <LOG>)`. When the end of the file is reached, <LOG> returns undef, undefining $_. My snippets demonstrate this behaviour.	[reply] [d/l] [select]
Re^3: Side Effects by goodgulf (Initiate) on Apr 07, 2005 at 20:18 UTC
OK, here's what I'm thinking. `$Body .= $_ while <LOG>;` [download] is equivalent to `while ($_ = <LOG>) { $Body .= $_; }` [download] Therefore, $_ is both an L-value and an R-value. If we assume for a moment that $_ is somehow an alias for $g::TaskList, then it should be correct that for each iteration of the while loop, $_ (and thus $g::TaskList) will get assigned the next line from <LOG>. At the end of the while loop, what would happen to $_? Would it get undef'd? Would this undef any aliases? So, let's assume all this is true... Here is my code path: In Localize.pm: (SNIP) foreach my $lang (@g::LanguageList) { $g::Language = $lang; unless (isExcludedBuild($g::Product, $g::Architecture, $g::Flavor) + \|\| $g::Architecture =~ /ia64/i \|\| $g::Language =~ /usa/i) { (SNIP) &BuildBlah(); } } (SNIP) In Utilities.pm: sub BuildBlah { my $FlavorVariation = shift; (snip) if(FileExists("Blah")) { SendMail($g::MailPurpose{"BuildFailed"}, "Blah", $FlavorVariat +ion); } (snip) } In Reporting.pm: sub SendMail { my $Purpose = shift; my @OptionalParameters = @_; my $Body = ""; (snip) $Body = "<html> <body background=nothing bgproperties=\"fixed\">"; if($Purpose eq $g::MailPurpose{BuildNominated}) { (snip) } elsif($Purpose eq $g::MailPurpose{BuildFailed}) { (snip) open(LOG, "$LogDir\\$g::BuildErrFile"); while (my $line = <LOG>) { $Body .= "${line}<br>"; } close LOG; } [download] Although I've snipped a lot of code, this follows the execution path, and you can see that $g::TaskList is never mentioned going way way back up through several loops, modules and subroutine calls. Do I need to backtrack further? I would think the foreach above would wipe out any alias of $_ to TaskList that might have been set higher up in the stack trace.	[reply] [d/l] [select]
Re: Side Effects by dragonchild (Archbishop) on Apr 07, 2005 at 00:18 UTC
Another random, and very unlikely, possibility is that some sick twisted soul has $Body as a tied variable to something whose STORE does nasty things to `$g::TaskList[0]`. Unlikely, but possible. My wife's blog	[reply] [d/l]
Re: Side Effects by tlm (Prior) on Apr 06, 2005 at 23:53 UTC
I don't see the problem immediately, but I can tell you that the working version is far from identical to the original. In ~~this~~ the working version the ~~loop~~ variable holding the last read input line is a lexical scalar, while in the buggy version it is a global. Update: I still can't figure it out; I need to see more code, preferably something I can run and reproduce the error with. Also, I fixed the wording of my original reply. the lowliest monk	[reply]