Re: [PATCH] fold strncmp of unterminated arrays (PR 92501)

[Jeff Law <law at redhat dot com>]

On 11/14/19 10:11 AM, Martin Sebor wrote:
> Adding tests for unsafe uses of unterminated constant char arrays
> in string functions exposed the limitation in strncmp folding
> described in PR 92501: GCC only folds strncmp calls involving
> nul-terminated constant strings.
> 
> The attached patch improves the folder to also handle unterminated
> constant character arrays.  This capability is in turn relied on
> for the dependable detection of unsafe uses of unterminated arrays
> in strncpy, a patch for which I'm about to post separately.
> 
> Tested on x86_64-linux.
> 
> Martin
> 
> gcc-92501.diff
> 
> PR tree-optimization/92501 - strncmp with constant unterminated arrays not folded
> 
> gcc/testsuite/ChangeLog:
> 
> 	PR tree-optimization/92501
> 	* gcc.dg/strcmpopt_7.c: New test.
> 
> gcc/ChangeLog:
> 
> 	PR tree-optimization/92501
> 	* gimple-fold.c ((gimple_fold_builtin_string_compare): Let strncmp
> 	handle unterminated arrays.  Rename local variables for clarity.
> 
> Index: gcc/gimple-fold.c
> ===================================================================
> --- gcc/gimple-fold.c	(revision 278253)
> +++ gcc/gimple-fold.c	(working copy)
> @@ -2361,9 +2361,32 @@ gimple_fold_builtin_string_compare (gimple_stmt_it
>        return true;
>      }
>  
> -  const char *p1 = c_getstr (str1);
> -  const char *p2 = c_getstr (str2);
> +  /* Initially set to the number of characters, including the terminating
> +     nul if each array has one.   Nx == strnlen(Sx, Nx) implies that
> +     the array is not terminated by a nul.
> +     For nul-terminated strings then adjusted to their length.  */
> +  unsigned HOST_WIDE_INT len1 = HOST_WIDE_INT_MAX, len2 = len1;
> +  const char *p1 = c_getstr (str1, &len1);
> +  const char *p2 = c_getstr (str2, &len2);
>  
> +  /* The position of the terminting nul character if one exists, otherwise
terminting/terminating/


OK with the nit fixed.
jeff