Re: Different behaviour of 'stat' function after assignment to $! (errno).

If 0 is assigned to $! before the first stat, the first failed stat does not set the $!.
But it does (as the numerous examples above show). The failed stat fills the Numeric Value of $!, and using $! in your code reads it (Devel::Peek::Dump shows the flags of the var not including pIOK, so the Integer Value is not used). However using -$! forces perl to use the Integer Value instead, which still contains your assigned value.(This was already pointed out in Re^3: Different behaviour of 'stat' function after assignment to $! (errno). by Anonymous Monk, but I thought I might explain it with a bit more words).
This could be considered unwanted behaviour, and you could opt to patch perl to set or clear the IV on error.

On the other hand, you could stop using -$! in your checks :)

Comment on Re: Different behaviour of 'stat' function after assignment to $! (errno).

Replies are listed 'Best First'.
Re^2: Different behaviour of 'stat' function after assignment to $! (errno). by jpl (Monk) on May 24, 2011 at 13:20 UTC
It's a little deeper than that. `use Devel::Peek; $! = 0.0; Dump($!); stat("non_existing_file"); Dump($!); die "died at first stat" if -$!; stat("non_existing_file"); Dump($!); die "died at second stat" if -$!;` [download] now yields `SV = PVMG(0x82c690) at 0x803478 REFCNT = 1 FLAGS = (GMG,SMG,pIOK,pNOK) IV = 0 NV = 0 PV = 0 MAGIC = 0x804900 MG_VIRTUAL = &PL_vtbl_sv MG_TYPE = PERL_MAGIC_sv(\0) MG_OBJ = 0x803460 MG_LEN = 1 MG_PTR = 0x8032d0 "!" SV = PVMG(0x82c690) at 0x803478 REFCNT = 1 FLAGS = (GMG,SMG,pIOK,pNOK) IV = 0 NV = 0 PV = 0 MAGIC = 0x804900 MG_VIRTUAL = &PL_vtbl_sv MG_TYPE = PERL_MAGIC_sv(\0) MG_OBJ = 0x803460 MG_LEN = 1 MG_PTR = 0x8032d0 "!" died at first stat at bug.pl line 6.` [download] pNOK is now set, but the value hasn't been updated by stat itself. It's only when $! is referenced that the updated value materializes. I certainly wouldn't recommend using -$!, but perlvar says If used numerically, yields the current value of the C "errno" variable, or in other words, if a system or library call fails, it sets this variable. This means that the value of $! is meaningful only immediately after a failure: so one can be excused for doing numerical things. It also says You can assign a number to $! to set errno if, for instance, you want "$!" to return the string for error n, or you want to set the exit value for the die() operator. and having to assign a non-integer value to avoid surprises doesn't fall in my category of "do what I mean".	[reply] [d/l] [select]
Re^3: Different behaviour of 'stat' function after assignment to $! (errno). by ikegami (Patriarch) on May 24, 2011 at 15:56 UTC
pNOK is now set, but the value hasn't been updated by stat itself. It's only when $! is referenced that the updated value materializes. Correct. It's set by SvGETMAGIC, which is called when the value of a magic SV is needed. The state of pNOK and the like should be completely ignored by `-$!` since negation should call SvGETMAGIC, but negation isn't calling SvGETMAGIC in the OP's version of Perl.	[reply] [d/l]
Re^3: Different behaviour of 'stat' function after assignment to $! (errno). by Anonymous Monk on May 24, 2011 at 13:31 UTC
and having to assign a non-integer value to avoid surprises doesn't fall in my category of "do what I mean". :) Um, do not assign anything to $!, unless you're curious about what types of error messages your system can generate -- errno is there for the system to assign to	[reply]
Re^2: Different behaviour of 'stat' function after assignment to $! (errno). by muxxum (Initiate) on May 24, 2011 at 12:52 UTC
This was a toy example. -$! is returned from a perl callback for which it makes sense to return a negated errno back to C. Unfortunately, strange bugs started to pop up under FreeBSD, and this issue could be related to it. Thanks everyone for helping.	[reply]