[libc-commits] [libc] [libc]: Implement strfrom* functions and utils (PR #85438)

[Michael Jones via libc-commits]

================
@@ -0,0 +1,134 @@
+//===-- Implementation header for strfromx() utilitites -------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+// According to the C23 standard, any input character sequences except a
+// precision specifier and the usual floating point formats, namely
+// %{a,A,e,E,f,F,g,G}, are not allowed and any code that does otherwise results
+// in undefined behaviour; which in this case is that the buffer string is
+// simply populated with the format string. The case of the value of buffer size
+// being 0 or the input being NULL should be handled in the calling function
+// (strfromf, strfromd, strfroml) itself.
+
+#ifndef LLVM_LIBC_SRC_STDLIB_STRFROM_UTIL_H
+#define LLVM_LIBC_SRC_STDLIB_STRFROM_UTIL_H
+
+#include "src/__support/str_to_integer.h"
+#include "src/stdio/printf_core/converter_atlas.h"
+#include "src/stdio/printf_core/core_structs.h"
+#include "src/stdio/printf_core/writer.h"
+
+#include <stddef.h>
+
+namespace LIBC_NAMESPACE::internal {
+
+template <typename T> struct type_of {
+  using type = T;
+};
+
+template <> struct type_of<float> {
+  using type = fputil::FPBits<float>::StorageType;
+};
+template <> struct type_of<double> {
+  using type = fputil::FPBits<double>::StorageType;
+};
+template <> struct type_of<long double> {
+  using type = fputil::FPBits<long double>::StorageType;
+};
+
+template <typename T> using type_of_v = typename type_of<T>::type;
+
+template <typename T>
+printf_core::FormatSection parse_format_string(const char *__restrict format,
+                                               T fp) {
+  printf_core::FormatSection section;
+  size_t cur_pos = 0;
+
+  if (format[cur_pos] == '%') {
+    section.has_conv = true;
+    ++cur_pos;
+
+    // handle precision
+    section.precision = -1;
+    if (format[cur_pos] == '.') {
+      ++cur_pos;
+      section.precision = 0;
+
+      // The standard does not allow the '*' (asterisk) operator for strfromx()
+      // functions
+      if (internal::isdigit(format[cur_pos])) {
+        auto result = internal::strtointeger<int>(format + cur_pos, 10);
+        section.precision += result.value;
+        cur_pos += result.parsed_len;
+      }
+    }
+
+    section.conv_name = format[cur_pos];
+    switch (format[cur_pos]) {
+    case '%':
+      section.has_conv = true;
+      break;
+    case 'a':
+    case 'A':
+    case 'e':
+    case 'E':
+    case 'f':
+    case 'F':
+    case 'g':
+    case 'G':
+      section.conv_val_raw = cpp::bit_cast<type_of_v<T>>(fp);
+      break;
+    default:
+      // error out, invalid format specifier
+      section.has_conv = false;
+      while (format[cur_pos] != '\0')
+        ++cur_pos;
+      break;
+    }
+
+    if (format[cur_pos] != '\0')
+      ++cur_pos;
+  } else {
+    section.has_conv = false;
+    // We are looking for exactly one section, so no more '%'
+    while (format[cur_pos] != '\0')
+      ++cur_pos;
+  }
+
+  section.raw_string = {format, cur_pos};
+  return section;
+}
+
+int convert(const printf_core::FormatSection &section,
+            printf_core::Writer *writer) {
+  if (!section.has_conv)
+    return writer->write(section.raw_string);
+
+  switch (section.conv_name) {
+  case '%':
+    return writer->write("%");
+  case 'f':
+  case 'F':
+    return convert_float_decimal(writer, section);
----------------
michaelrj-google wrote:

I see what the problem you're running into is: `convert_float_decimal` selects the type based on the `FormatSection` it's passed (see: https://github.com/llvm/llvm-project/blob/main/libc/src/stdio/printf_core/float_dec_converter.h#L1112)

The easiest way to solve this problem would be to copy the code to check for inf/nan and then call `convert_float_decimal_typed<T>` (and similar below)

At some point those functions should be refactored to have one `write_if_inf_or_nan` function that they all call, but you don't have to do that if you don't want to.

https://github.com/llvm/llvm-project/pull/85438