Re^7: Changing array by changing $

There's a mechanism to cope with:

  func('Hello Sailor') ;
[download]

which might be extended to passing of aliases, under suitable strictness...

You are right. This particular trap is deeper, darker and altogether more ghastly than I thought !

Comment on Re^7: Changing array by changing $_? Download Code

Replies are listed 'Best First'.
Re^8: Changing array by changing $_? by ikegami (Patriarch) on Oct 13, 2008 at 18:56 UTC
Said mechanism is to make the SV readonly. `>perl -MDevel::Peek -e"Dump('Hello Sailor')" SV = PV(0x228e68) at 0x1831670 REFCNT = 1 FLAGS = (PADBUSY,PADTMP,POK,READONLY,pPOK) ^^^^^^^^ PV = 0x1833f3c "Hello Sailor"\0 CUR = 12 LEN = 16` [download] Obviously that mechanism won't help you here. In fact, a flag alone won't work since you'd need to know how many times a variable has been aliased to know when to remove the flag. (Upd: I'm not sure about the preceding statement. Can't think straight right now. ) You could add magic to the variable like taint does. `>perl -T -MDevel::Peek -e"Dump($ARGV[0])" foo SV = PVMG(0x1822dcc) at 0x226ba0 REFCNT = 1 FLAGS = (GMG,SMG,pPOK) IV = 0 NV = 0 PV = 0x182d324 "foo"\0 CUR = 3 LEN = 4 MAGIC = 0x182d33c MG_VIRTUAL = &PL_vtbl_taint MG_TYPE = PERL_MAGIC_taint(t) MG_LEN = 1` [download] But now you're getting into costly.	[reply] [d/l] [select]
Re^9: Changing array by changing $_? by gone2015 (Deacon) on Oct 13, 2008 at 22:07 UTC
OK, the penny is dropping... So the `@_` is an array of SV*. To distinguish aliases as arguments requires the pointer, not the target, to be marked -- it's only assignments via the alias that should be trapped. That would add overhead. Mind you, the compiler knows when it's assigning to `$_[x]`, and only then does it need to worry about copies of aliases -- I think the propagation of aliases is limited -- so I guess the overhead could be contained... assuming one could figure out a way of passing the information with the pointer. Hmmm. I'm starting to feel queasy. Well, nothing's perfect... if the 99% cases (lexically inside the `foreach`/`map`/etc) were trapped at compile-time it would be a step in the right direction.	[reply] [d/l] [select]