My somewhat wild conjecture is that this is related to how line numbers are reported incorrectly for elsif code (the number of the line of the if is reported instead).
The broken line numbering is clearly a bug and justifies more effort be marshaled to fix it than the debugger issue.
Further, I conjecture that both are somewhat affected by the Perl4 optimization where a suitable if/elsif/else chain is converted into an efficient 'switch statement' (computed goto).
| [reply] |
Further research...given this:
#!/your/perl/here
use strict;
use warnings;
my $rand = rand;
if ( $rand < rand )
{
print "if1: $rand\n";
}
elsif ( $rand < rand )
{
print "elsif2: $rand\n";
}
else
{
print "else3: $rand\n";
}
works fine. Changing line 11 to:
elsif ( ( $rand < rand ) { print "elsif2: $rand\n"; }
gives the error:
syntax error at C:\if_elsif.pl line 12, near ")
{"
But using this version instead:
elsif ( $rand < _rand ) { print "elsif2: $rand\n"; }
gives the compile error:
Bareword "_rand" not allowed while "strict subs" in use at C:\if_elsif
+.pl line 7.
Given that you can chain elsifs together indefinitely, it is possible to cause an error at any line after line 7, but still be pointed to line 7.
You may be wondering what happens if use strict is turned off - it just generates a warning instead:
Argument "_rand" isn't numeric in numeric lt (<) at C:\Perl\perlmonks\
+if_elsif.pl line 7.
Turning off warnings then doesn't complain. I was unable to generate a message that pointed to some line besides the line of the actual error +/-1. Only turning on strict and warnings caused misleading messages.
So it appears (from minimial empirical evidence) that syntax errors generate mostly correct line numbers, but strict and warnings do not, probably because information about the structure of the if/elsif/else has been removed from the [insert appropriate entity name here]
However, if another if/elsif/else is nested in one of the if blocks, it's if entry point is retained (but it's elsif/else info is not).
The remaining question I have is:
What information is thrown out, and at what stage of the compilation/execution?
[The original question, "Why doesn't the debugger pause on elsif conditionals?", seems to hinge on the answer to this new question.]
-QM
--
Quantum Mechanics: The dreams stuff is made of
| [reply] [d/l] [select] |
What information is thrown out, and at what stage of the compilation/execution?
IIRC, after compilation, line numbers are only stored for the beginning of each statement. Errors generated by the lexer or parser (e.g. "syntax error") use the actual line being parsed, but any other perl error is going to show the beginning line of the statement. Does that help?
| [reply] |
/me points at "Larry's Head".
| [reply] |
I'm no expert on this but I goaded a few experts on this and in the end I don't think they came up with any good reasons why 'elsif' shouldn't just be declared "a statement" as far as opcode generation is concerned so that run-time line numbers can be correct and the debugger can stop on such lines.
I expected that this would find its way to p5p and someone would post some evidence of that to this thread. Maybe this goading will make that happen. If not, you might want to escalate to p5p yourself (after first searching the archive to see if it has already been discussed).
| [reply] |
This involves making the optimizer optimize less for a case that it already did well enough in. This new optimize-by-doing-less would only be appropriate during -d debugging and then might be considered a source of heisenbugs.
| [reply] |