[Bioperl-l] Bio::SeqIO::new possible wierdness

[Donald G. Jackson]

Personally, I like the fall-back but agree that $ARGV[0] shouldn't be it.
I'd suggest STDIN - if somebody calls new without a file/handle I think 
they're more likely to be reading.  OTOH, guessing format woud be tough.
At the very least a warning would be appropriate, perhaps indicating the 
course of action.

For xml handlers we can check the dtd and throw an error.  I will modify 
my SeqIO::tinyseq::tinyseqHandler to do so.

Don Jackson



Peter van Heusden wrote:

> My review of the Bio::SeqIO::new method shows the following behaviour:
>
> Missing both –file and –fh arguments: falls back to using $ARGV[0] 
> (the first command line argument) as sequence filename. If this fails, 
> gives an exception about ‘Unknown format’.
> -file argument (without –fh argument):
> · given, but file unreadable: throws exception
> · undefined: reads $ARGV[0], as above.
> -fh argument (without –file argument):
> · given, but not a filehandle: gives exception
> · given, but an invalid filehandle (not open): gives exception
> · undefined: reads $ARGV[0], as above.
> -format argument: if the sequence file doesn’t correspond to the given 
> format, some parsers give an error (e.g. EMBL), while others do not 
> (GenBank), instead silently give wrong results.
> -format argument without –file argument: Silently creates a SeqIO 
> object which writes to STDOUT.
>
> I don't think that this $ARGV[0] shortcut should be in there - it 
> causes unnecessary potential confusion. Imagine a situation where -fh 
> or -file is specified (using a variable), but that variable somehow 
> does not get defined. In that case, the $ARGV[0] fallback behaviour 
> would be used, which might lead to a non-obvious error behaviour.
>
> I'd like to propose that either -file or -fh should be specified, 
> otherwise an exception is thrown. While I'm about it, I'm thinking of 
> migrating the exceptions to the new 'typed exceptions' that BioPerl 
> now provides - is there any consensus on exception type names?
>
> Peter
> _______________________________________________
> Bioperl-l mailing list
> Bioperl-l at portal.open-bio.org
> http://portal.open-bio.org/mailman/listinfo/bioperl-l
>