On Sat, Oct 29, 2005 at 10:59:00AM +0100, Peter Stephenson wrote:
> you will need to be careful since those "int" definitions for the
> variables now marked as ZLE_CHAR_T have been around a very long time
> and it's not necessarily clear if they will hold an EOF or not, though
> if they do they should certainly now be ZLE_INT_T.

I looked at all the ZLE_CHAR_T variables while making my patch in order
to check for size differences (which is why a couple spots got changed
to use ZLE_INT_T).

One other potential problem area is less-than and greater-than
comparisons of character values now that the char may be signed.  I just
looked through the code to check for problems, and while I didn't see
any bugs, I did see a couple things that could be improved (such as
using idigit() instead of doing a range-comparison between '0' & '9').

Yet another potential problem area is calling functions such as
islower() with a signed character.  I looked through the code, and found
quite a few problems not related to my patch, including the calling of
some non-wide isTYPE() functions on wide characters in the zleline
array.  Another thing I noticed was that the wide version of ZC_iblank()
was calling iswspace(), which would return true for a newline, unlike
the normal iblank(), which does not (that is reserved for inblank()).

So, attached is a patch that fixes the problems I discovered:

I decided to add some new iTYPE() macros, namely iascii(), ilower(),
iprint(), and iupper().  I didn't modify inittyptab() to try to define
these values (for a number of reasons).  Instead, they ensure that
STOUC() gets called on the arg before we call the system's version of
these functions (aside: we could switch some of the existing macros over
to this method, such as idigit(), and free up some bits in the typtab[]
array).

I also added any missing ZC_iTYPE macros in zle.h, just so they'd all be
there.  The wide-char version of ZC_iblank now calls a new function,
wcsiblank(), that properly excludes '\n' from returning true.  However,
both ZC_iblank and ZC_inblank still make use of iswspace(), so they
return true for a carriage return ('\r') and a vertical tab ('\v') where
iblank and inblank do not.  I didn't try to change that one way or the
other.

Comments welcomed.

..wayne..