If a user does use v7 or -M7 under this proposal, part of what they have opted into is for their source code to be decoded to characters. Changing the behavior of the global STDIN and STDOUT handles in a reasonable way is unfortunately impossible, but you can already do that with -CS if you accept the consequences.
My oneliners often use ojo which already enables the utf8 pragma and it's operated as expected. Data that flows from STDIN to STDOUT would be unchanged by this, though you already needed to use -CS or appropriate decoding and encoding if you want to operate on it as text. Unfortunately there is no way around learning how and when character encoding occurs if you want to interact with text as bytes.
The proposal here, though, defines what someone opts into.
All the other pieces of your proposal seem, at least from my own vantage point, to be “easy wins”. Auto-decode without an auto-encode, though, seems ripe for subtle misuse. If Perl could somehow mark the PVs as decoded, and always trigger a warning or error on output, I’d be less concerned.
Forgive my ignorance, but why would enabling -CS by default be any less feasible than use utf8 by default?
I don't see how use utf8 is analogous to auto-encoding on STDOUT - the opposite is of course auto-decoding from STDIN. use utf8 is instead a lexical declaration of how the source code shall be interpreted.
The problem with -CS and any other application of layers to STDIN/STDOUT/STDERR is that the handles and any layers applied to them are global. So for example, it will cause Mojo::Log's encoded output to STDERR to be double-encoded. (This experiment was attempted in Perl 5.8.0 and failed miserably.)
If there were a variant of use utf8 that didn’t auto-decode strings in the source, I’d be much less concerned. But the issue I see with defaulting use utf8 to on is that it would break any code like this:
perl -e'print "épée"'
In fact, it’ll even break things like this:
my $text = utf8_decode("épée");
_send_to_dbus($text);
Ostensibly the goal of Perl 7 would be to define a set of defaults that only break “undesirable” practices. Changing the value of hard-coded strings in the source code seems likely to break a lot of things and thus deter people from using the new set of defaults.
"A variant of use utf8 that didn't auto-decode strings in the source" would be a no-op - that is the only thing use utf8 does.
I appreciate your opinion though I believe it would be more helpful to new code than harmful. The purpose of use v7 is of course not to blindly apply to existing code - as proposed, it will also break any code defining subroutine prototypes, for example.
Prototypes have been “gently discouraged” for some time, though, AFAIK. More so, I think, than writing new Perl without use utf8.
use utf8 seems the most disruptive of the changes you propose—disruptive insofar as that developers themselves would need to exercise especial care when writing new code or porting existing code. use v7 defined with use utf8 would be problematic where I work, for example, where strings are understood by default to be undecoded/binary/encoded. Whereas enabling strict/warnings/signatures will generate “loud”, easily-fixed breakages, breakages from auto-decode of strings in source seem likely to be subtler.
Anyhow … the appreciation of opinions is mutual. :) We’ll see what comes. Thanks!
For further actions, you may consider blocking this person and/or reporting abuse
We're a place where coders share, stay up-to-date and grow their careers.
If a user does
use v7or-M7under this proposal, part of what they have opted into is for their source code to be decoded to characters. Changing the behavior of the global STDIN and STDOUT handles in a reasonable way is unfortunately impossible, but you can already do that with-CSif you accept the consequences.My oneliners often use ojo which already enables the
utf8pragma and it's operated as expected. Data that flows from STDIN to STDOUT would be unchanged by this, though you already needed to use-CSor appropriate decoding and encoding if you want to operate on it as text. Unfortunately there is no way around learning how and when character encoding occurs if you want to interact with text as bytes.The proposal here, though, defines what someone opts into.
All the other pieces of your proposal seem, at least from my own vantage point, to be “easy wins”. Auto-decode without an auto-encode, though, seems ripe for subtle misuse. If Perl could somehow mark the PVs as decoded, and always trigger a warning or error on output, I’d be less concerned.
Forgive my ignorance, but why would enabling
-CSby default be any less feasible thanuse utf8by default?I don't see how
use utf8is analogous to auto-encoding on STDOUT - the opposite is of course auto-decoding from STDIN.use utf8is instead a lexical declaration of how the source code shall be interpreted.The problem with
-CSand any other application of layers to STDIN/STDOUT/STDERR is that the handles and any layers applied to them are global. So for example, it will cause Mojo::Log's encoded output to STDERR to be double-encoded. (This experiment was attempted in Perl 5.8.0 and failed miserably.)If there were a variant of
use utf8that didn’t auto-decode strings in the source, I’d be much less concerned. But the issue I see with defaultinguse utf8to on is that it would break any code like this:In fact, it’ll even break things like this:
Ostensibly the goal of Perl 7 would be to define a set of defaults that only break “undesirable” practices. Changing the value of hard-coded strings in the source code seems likely to break a lot of things and thus deter people from using the new set of defaults.
"A variant of
use utf8that didn't auto-decode strings in the source" would be a no-op - that is the only thinguse utf8does.I appreciate your opinion though I believe it would be more helpful to new code than harmful. The purpose of
use v7is of course not to blindly apply to existing code - as proposed, it will also break any code defining subroutine prototypes, for example.Prototypes have been “gently discouraged” for some time, though, AFAIK. More so, I think, than writing new Perl without
use utf8.use utf8seems the most disruptive of the changes you propose—disruptive insofar as that developers themselves would need to exercise especial care when writing new code or porting existing code.use v7defined withuse utf8would be problematic where I work, for example, where strings are understood by default to be undecoded/binary/encoded. Whereas enabling strict/warnings/signatures will generate “loud”, easily-fixed breakages, breakages from auto-decode of strings in source seem likely to be subtler.Anyhow … the appreciation of opinions is mutual. :) We’ll see what comes. Thanks!