FX,
You're likely correct that the creation of an expression for
c_null_char is overkill. I copied the code from gfc_conv_concat_op()
in trans-expr.cc to get the 'f_c_string(s) -> trim(s) // c_null_char'.
This was the motivation of creating a string. If I look at the
-fdump-tree-orginal for
s1 = 'abc '
s2 = f_c_string(s1)
I see (slightly edited to reduce noise)
_string_trim (&tlen.0, (void **) &tstr.1, 6, &s1);
D.4785 = (void *) __builtin_malloc (MAX_EXPR <(tlen.0 + 1), 1>);
pstr.2 = (character(kind=1) *) D.4785;
_concat_string (tlen.0+1, pstr.2, tlen.0, tstr.1, 1, &"z"[1]{lb: 1 sz: 1});
so the c_null_char (currently 'z') simply appears as a constant.
I keep your comment in mind when I polish the code.
--
steve
On Tue, Dec 10, 2024 at 11:15:10PM +0100, FX Coudert wrote:
> Hi Steve,
>
> I think it’d be simpler not to create a front-end expression for the
> c_null_char. Unless I’m mistaken, you only need it as an argument to pass to
> build_call_expr_loc (input_location, gfor_fndecl_concat_string, …). You can
> pass a pointer to the constant character, and the length is one. The variable
> for length would be build_int_cst (gfc_charlen_type_node, 1). For the pointer
> to the constant character, you would do:
>
> tree nulchar = gfc_character1_type_node(gfc_character1_type_node,
> “nulchar”);
> gfc_add_modify (&block, nulchar, build_zero_cst (gfc_character1_type_node));
> tree pnulchar = gfc_build_addr_expr (NULL_TREE, nulchar);
>
> Since the variable in this case is a known constant, there’s probably even a
> better way. I just don’t remember enough, but others may help?
>
> ---
>
> Regarding your question, you are right: you want to build the two code blocks
> (you already have the content), then use: fold_build3_loc (input_location,
> COND_EXPR, return_type, boolexpression, iftrue, iffalse)
>
> You will find some logic like that in gfc_conv_intrinsic_minmaxloc() near the
> comment "We switch to > or >= depending on the value of the BACK argument”.
> You create ifblock and elseblock, once you have the code inside them, you
> make them into trees with gfc_finish_block (becoming ifbody2 and elsebody2).
> Then:
>
> tmp = fold_build3_loc (input_location, COND_EXPR, logical_type_node,
> back, ifbody2, elsebody2);
>
>
> Hope this helps?
>
>
> Best,
> FX
--
Steve
