No Pause on Elsif in Debugger

[QM]

I've noticed recently that the debugger doesn't pause on elsifs, making it confusing to follow sometimes.

For instance, given this code:

#!/your/perl/here

use strict;
use warnings;

my $rand = rand;

if ( $rand < rand ) # 1
{
    print "if 1: $rand\n";
}
elsif ( $rand < rand ) # 2
{
    print "elsif 2: $rand\n";
}
elsif ( $rand < rand ) # 3
{
    print "elsif 3: $rand\n";
}
else
{
    print "else 4: $rand\n";
}

I get the following in a debugger session:

C:\Perl\perl\perlmonks>perl -d skip_elsif.pl

Loading DB routines from perl5db.pl version 1.19
Editor support available.

Enter h or `h h' for help, or `perldoc perldebug' for more help.

main::(skip_elsif.pl:6):        my $rand = rand;
  DB<1> s
main::(skip_elsif.pl:8):        if ( $rand < rand ) # 1
main::(skip_elsif.pl:9):        {
  DB<1> p $rand
0.63232421875
  DB<2>
main::(skip_elsif.pl:22):               print "else 4: $rand\n";
  DB<2>
else 4: 0.63232421875

I'm expecting the debugger to pause on each conditional, especially as I might want to inspect some of the variables before the next evaluation. [I wouldn't mind if it paused on elses also, but I supposed that's too much to ask?]

Why does it work this way?

[merlyn]

It "pauses" on the beginning of a statement. elsif is not the beginning of a statement.

[QM]

It "pauses" on the beginning of a statement. elsif is not the beginning of a statement.

Is that DWIM? What behavior do we want?

I expect that it will at least pause on the evaluation of interesting conditionals [elsif (0) being an example of an uninteresting one]

[merlyn]

It doesn't really matter if that's exactly what you want. It would take a major overhaul (major) of the interface between the execution engine and the debugger Perl code to get it to do anything else. It's a very fundamental design choice. It's just like you can't set a breakpoint to see the "middle" of a map operation, other than at a statement boundary in a called subroutine.

The debugger gets notified only as each statement is entered. That's it.

[pemungkah]

Of course, the Right Place to put the breakpoints is on the code *inside* the if/elsif/else - "right" being "you will actually stop somewhere near where you want to be". I've noticed similar annoying behavior with for loops at times. Just set your breakpoints inside and the problem goes away. This is the least intrusive solution; obviously, you can code to the debugger's limitations by doing things like

if ($cond1) {
  ...
}
else {
  if ($cond2) {
    ...
  }
  else {
    if ($cond3) {
      ..
    }
  }
}

and now you can break at every if. (That looks scarily like some of my old PL/1 programs...)

[qq]

Update: missed the point.

Its just following the control flow. Its not going to evaluate elsifs that it doesn't need to, because a prior condition passed. I think the rands in your statements are obscuring the issue.

if ( 1 ) {
  # debugger will see this, just like code
} elsif ( 1 ) {
  # won't see this - nothing will
} else {
  # nor this either
}

qq

[matija]

Um, you're missing something, too. Take the code:

if ( 0 ) {
    warn "a";
} elsif (1) {
    warn "b";
} else {
    warn "c";
}

You will see this execution:

main::(/tmp/foo:12):    if ( 0 ) {
  DB<1> n
main::(/tmp/foo:15):        warn "b";
  DB<1> n
b at /tmp/foo line 15.

So in an elsif chain, you always see just the first condition, not the first matching condition as your post states.

20040725 Edit by ysth: correct <code> to </code>

[CountZero]

Attention: Could it not be that the optimizer simply "optimizes away" the code it can never reach? This would not happen with the "rand" statements.

Anyhow it doen't matter as far as the debugger is concerned: the debugger only stops on "if"-part.

[CountZero]

I checked it in the graphical debugger of KOMODO and it stops only on the first "if" statement and then goes on to the statements after the "elseif", never stopping on these "elseif".

Even if you put a "breakpoint" on an "elseif", the debugger stops on the first "if".

It seems the whole of the "if -elseif - else" construct is indeed considered "one"