What happens with freadahead() and freadptr() after ungetc?
In theory it is possible that an implementation stores the bytes pushed-back
by ungetc in a location that freadahead() does not see. The implementations
that I tested don't do this, but the API should support this. I can see three
possible ways to support this situation:
(a) Change freadahead(), so that freadahead() after ungetc return 0.
(b) Change freadptr() so that it returns NULL after ungetc, even though
freadahead() > 0.
(c) Add a new API that tests whether the stream is current reading pushed-back
bytes.
(a) would break the closein.c code. It assumes that when freadahead() there
are no buffers in memory (except for the pushed-back bytes).
(c) may be difficult. I already had problems implementing it on glibc.
So this implements (b).
*** lib/freadptr.h.orig 2008-02-29 00:49:20.000000000 +0100
--- lib/freadptr.h 2008-02-28 23:32:46.000000000 +0100
***************
*** 23,30 ****
/* Assuming the stream STREAM is open for reading:
Return a pointer to the input buffer of STREAM.
! If freadahead (STREAM) > 0, the result is a pointer to freadahead (STREAM)
! bytes.
If freadahead (STREAM) == 0, the result is not usable; it may be NULL.
In this case, you should use getc (STREAM), fgetc (STREAM), or
fread (..., STREAM) to access the input from STREAM.
--- 23,31 ----
/* Assuming the stream STREAM is open for reading:
Return a pointer to the input buffer of STREAM.
! If freadahead (STREAM) > 0, the result is either a pointer to
! freadahead (STREAM) bytes, or NULL. The latter case can happen after
! use of 'ungetc (..., STREAM)'.
If freadahead (STREAM) == 0, the result is not usable; it may be NULL.
In this case, you should use getc (STREAM), fgetc (STREAM), or
fread (..., STREAM) to access the input from STREAM.
