[libc][z/OS] Remove ASCII trick to fix EBDIC std::from_char

[zibi2]

This PR will fix the following lit in all EBCDIC variations on z/OS:
std/utilities/charconv/charconv.from.chars/floating_point.pass.cpp

The trick to test for e and E is working only in ASCII.
The fix is to simply test for both lower and upper case exponent letter e and E respectfully.

[llvmbot]

@llvm/pr-subscribers-libc

Author: Zibi Sarbinowski (zibi2)

Changes

This PR will fix the following lit in all EBCDIC variations on z/OS:
std/utilities/charconv/charconv.from.chars/floating_point.pass.cpp

The trick to test for e and E is working only in ASCII.
The fix is to simply test for both lower and upper case exponent letter e and E respectfully.

---

Full diff: https://github.com/llvm/llvm-project/pull/116078.diff

1 Files Affected:

• (modified) libc/src/__support/high_precision_decimal.h (+2-1)

diff --git a/libc/src/__support/high_precision_decimal.h b/libc/src/__support/high_precision_decimal.h
index ac11649d1d1686..20088d6d797910 100644
--- a/libc/src/__support/high_precision_decimal.h
+++ b/libc/src/__support/high_precision_decimal.h
@@ -350,7 +350,8 @@ class HighPrecisionDecimal {
     if (!saw_dot)
       this->decimal_point = total_digits;
 
-    if (num_cur < num_len && ((num_string[num_cur] | 32) == 'e')) {
+    if (num_cur < num_len &&
+        (num_string[num_cur] == 'e' || num_string[num_cur] == 'E')) {
       ++num_cur;
       if (isdigit(num_string[num_cur]) || num_string[num_cur] == '+' ||
           num_string[num_cur] == '-') {

[abhina-sree]

LGTM

[michaelrj-google]

Thanks for catching this, I'll see if there's anywhere else that could use this sort of cleanup.

[zibi2]

CI failed twice at unrelated generate_test_report.py step. I'm going to merge this now.

[nickdesaulniers]

@zibi2 what's a target triple we can use to test EBDIC? --target=s390x-???

@michaelrj-google is making more related changes in #110574.

[perry-ca]

(Zibi's off for a couple days. I'll answer in the meantime)
You don't really need to use a particular target triple. To compile something in EBCDIC you'll need:

1. the -fexec-charset option (see Create a CharSetConverter class with both iconv and icu support #74516 & Enable fexec-charset option abhina-sree/llvm-project#1) or #pragma convert (follow on to these)
2. A way to got through the preprocessing paths for defined(__MVS__) && !defined(__NATIVE_ASCII_F). This will then depend on macros or values defined in the z/OS headers (eg. static const mask space = __ISSPACE;)

This would be a specific build of libc++ for EBCDIC. We haven't tried enabling EBCDIC on other platforms. In theory it should work.