`mrb_read_float()` converts `"0.3"` with a small error compared to `strtod()`

[dearblue]

`mrb_read_float()` converts `"0.3"` with a small error compared to `strtod()`

[dearblue]

For information, it seems also to have the same problem with the original, https://github.com/mattn/strtod.
Adding values in 0.1 intervals from 0.2 to 0.9 and reporting only failures, the results are shown below.

%  cc strtod.c -lm && ./a.out
CASE: .3
  ERR: .3, No error: 0
    E1 0.300000, 0.3, , 0
    E2 0.300000, 0.3, , 0
different value

CASE: .6
  ERR: .6, No error: 0
    E1 0.600000, 0.6, , 0
    E2 0.600000, 0.6, , 0
different value

CASE: .7
  ERR: .7, No error: 0
    E1 0.700000, 0.7, , 0
    E2 0.700000, 0.7, , 0
different value

CASE: 2.2250738585072010e-308
  ERR: 2.2250738585072010e-308, Result too large
    E1 0.000000, 0, , 34
    E2 0.000000, 2.22507e-308, , 34
different value

CASE: 2.2250738585072011e-308
  ERR: 2.2250738585072011e-308, Result too large
    E1 0.000000, 0, , 34
    E2 0.000000, 2.22507e-308, , 34
different value

patch for test of strtod.c

diff --git a/strtod.c b/strtod.c
index bd5378b..42bb574 100644
--- a/strtod.c
+++ b/strtod.c
@@ -148,8 +148,6 @@ test(char* str)
     char *e1, *e2;
     int x1, x2;
 
-    printf("CASE: %s\n", str);
-
     errno = 0;
     e1 = NULL;
     d1 = vim_strtod(str, &e1);
@@ -161,22 +159,29 @@ test(char* str)
     x2 = errno;
 
     if (d1 != d2 || e1 != e2 || x1 != x2) {
+        printf("CASE: %s\n", str);
 	printf("  ERR: %s, %s\n", str, strerror(errno));
 	printf("    E1 %f, %g, %s, %d\n", d1, d1, e1 ? e1 : "", x1);
 	printf("    E2 %f, %g, %s, %d\n", d2, d2, e2 ? e2 : "", x2);
        if (d1 != d2) puts("different value");
        if (e1 != e2) puts("different end position");
        if (x1 != x2) puts("different errno");
-    } else {
-	printf("  SUCCESS [%f][%s]: %s\n", d1, e1 ? e1 : "", strerror(errno));
+        printf("\n");
     }
-    printf("\n");
 }
 
     int
 main()
 {
     test(".1");
+    test(".2");
+    test(".3");
+    test(".4");
+    test(".5");
+    test(".6");
+    test(".7");
+    test(".8");
+    test(".9");
     test("  .");
     test("  1.2e3");
     test(" +1.2e3");

[matz]

For several reasons, I recommend mruby-marshal to use mrb_read_float() instead of strtod().

• strtod() uses locale which is fragile (for example we need to use 0,3 not 0.3 for some locale)
• it is very hard to implement bit-to-bit equality with independent string-to-float functions

For the same reason, I recommend to use mrb_float_to_str() instead of printf().

[dearblue]

Understood. Thank you very much.
I tried that too, but it's not possible.

I think it makes sense to write it in doc/limitations.md as it is now, so I'll update that one.