[PATCH 2/3] ungetc: Guarantee single char pushback

[Siddhesh Poyarekar]

On 2024-11-29 02:02, Florian Weimer wrote:
> * Maciej W. Rozycki:
> 
>> The structure is fully packed and there's no padding in the legacy part
>> I'm afraid, and I guess it's the case for the majority of 32-bit targets,
>> which don't imply alignment beyond 4 bytes for pointers.  Have I missed
>> anything?
> 
> No, you are right.  So the extension area has to be used.  I think we
> should fail ungetc if the malloc call fails and the extension area is
> unavailable, instead of resorting to heroic bit-stuffing.

So basically the original struct layout I had proposed plus this in 
genops.c should do it?

diff --git a/libio/genops.c b/libio/genops.c
index d7e35e67d5..5b8bb52ec3 100644
--- a/libio/genops.c
+++ b/libio/genops.c
@@ -1002,7 +1002,14 @@ _IO_default_pbackfail (FILE *fp, int c)
               int backup_size = 128;
               char *bbuf = (char *) malloc (backup_size);
               if (bbuf == NULL)
-               return EOF;
+               {
+#if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_1)
+                 if (__glibc_unlikely (_IO_vtable_offset (fp) != 0))
+                   return EOF;
+#endif
+                 bbuf = fp->_short_backupbuf;
+                 backup_size = 1;
+               }
               fp->_IO_save_base = bbuf;
               fp->_IO_save_end = fp->_IO_save_base + backup_size;
               fp->_IO_backup_base = fp->_IO_save_end;