Mail Archives: cygwin-apps/2001/04/22/18:52:15
On Mon, Apr 23, 2001 at 12:36:23AM +0200, Corinna Vinschen wrote:
>On Sun, Apr 22, 2001 at 09:25:00AM -0400, Earnie Boyd wrote:
>> Corinna Vinschen wrote:
>> >
>> > On Sat, Apr 21, 2001 at 01:19:57PM -0400, Christopher Faylor wrote:
>> > > On Sat, Apr 21, 2001 at 08:25:56PM +0400, egor duda wrote:
>> > > >ash stdin mode is set to O_TEXT, so it's impossible to checkin/chechout
>> > > >binary files from cvs when CVS_RSH is sh script.
>> > > >
>> > > >maybe O_TEXT should be set only in interactive mode? or shouldn't be
>> > > >set at all?
>> > >
>> > > Corinna and I have discussed this in the past. The problem is that the
>> > > 'read' command should always be text mode whether it is interactive or
>> > > not. Also, if sh is reading a shell script via:
>> > >
>> > > sh < foo
>> > >
>> > > the input should be in O_TEXT mode.
>> > >
>> > > But then, if ash is reading from a pipe, stdin should, IMO, be in binary
>> > > mode.
>> >
>> > That's really tricky:
>> >
>> > sh < foo O_TEXT?
>> > cat foo | sh O_BINARY?
>> >
>>
>> Correct, and from what I remember of the bash code it knows when it's
>> about to read/write from/to a pipe/redirect. Wouldn't you just use the
>> appropriate setmode on the duplicated file handle?
>
>I'm not sure if I understand you right. Do you mean sth. like that:
>
> struct stat s;
> fstat(STDIN_FILENO, &s);
> if (S_ISCHR(s.st_mode) || S_ISREG(s.st_mode))
> setmode(STDIN_FILENO, O_TEXT);
> else
> setmode(STDIN_FILENO, O_BINARY);
Actually, the 'cat foo | sh' example should still default to text mode.
sh shouldn't interpret CRs when a pipe is opened.
I think what is really needed here is for sh to not use setmode at all.
It should probably set its stdin to binary and do the CR stripping itself.
cgf
- Raw text -