[flang-commits] [flang] [llvm] [flang][flang-rt][F202X][Issue #178494] Implement F202X leading-zero … (PR #183500)
via flang-commits
flang-commits at lists.llvm.org
Thu Feb 26 03:39:23 PST 2026
https://github.com/laoshd created https://github.com/llvm/llvm-project/pull/183500
…control edit descriptors LZ, LZS, and LZP for formatted output (F, E, D, and G editing): LZ: processor-dependent (default, flang prints leading zero); LZS: suppress the optional leading zero before the decimal point; LZP: print the optional leading zero before the decimal point. Changes span the source parser, compile-time format validator, runtime format processing, and runtime output formatting. Includes semantic test (io16.f90) and documentation updates.
>From e2922a19918f526805dcd58205d770c14c04abdf Mon Sep 17 00:00:00 2001
From: Shandong Lao <shandong.lao at hpe.com>
Date: Thu, 26 Feb 2026 05:03:53 -0600
Subject: [PATCH] [flang][flang-rt][F202X][Issue #178494] Implement F202X
leading-zero control edit descriptors LZ, LZS, and LZP for formatted output
(F, E, D, and G editing): LZ: processor-dependent (default, flang prints
leading zero); LZS: suppress the optional leading zero before the decimal
point; LZP: print the optional leading zero before the decimal point. Changes
span the source parser, compile-time format validator, runtime format
processing, and runtime output formatting. Includes semantic test (io16.f90)
and documentation updates.
---
.../flang-rt/runtime/format-implementation.h | 29 +++++-
flang-rt/include/flang-rt/runtime/format.h | 7 ++
flang-rt/lib/runtime/edit-output.cpp | 47 +++++++--
flang/docs/F202X.md | 9 ++
flang/docs/FortranStandardsSupport.md | 2 +-
flang/include/flang/Common/format.h | 45 ++++++++-
.../flang/Parser/format-specification.h | 3 +
flang/lib/Parser/io-parsers.cpp | 9 +-
flang/lib/Parser/unparse.cpp | 3 +
flang/test/Semantics/io16.f90 | 99 +++++++++++++++++++
10 files changed, 235 insertions(+), 18 deletions(-)
create mode 100644 flang/test/Semantics/io16.f90
diff --git a/flang-rt/include/flang-rt/runtime/format-implementation.h b/flang-rt/include/flang-rt/runtime/format-implementation.h
index d510adbb5ba46..8b341def2b3ce 100644
--- a/flang-rt/include/flang-rt/runtime/format-implementation.h
+++ b/flang-rt/include/flang-rt/runtime/format-implementation.h
@@ -193,7 +193,7 @@ static RT_API_ATTRS bool AbsoluteTabbing(CONTEXT &context, int n) {
template <typename CONTEXT>
static RT_API_ATTRS void HandleControl(
- CONTEXT &context, char ch, char next, int n) {
+ CONTEXT &context, char ch, char next, int n, char next2 = '\0') {
MutableModes &modes{context.mutableModes()};
switch (ch) {
case 'B':
@@ -251,6 +251,21 @@ static RT_API_ATTRS void HandleControl(
return;
}
break;
+ case 'L':
+ if (next == 'Z') {
+ if (next2 == 'S') {
+ // LZS - suppress leading zeros
+ modes.leadingZero = MutableModes::LeadingZeroMode::Suppress;
+ } else if (next2 == 'P') {
+ // LZP - print leading zero
+ modes.leadingZero = MutableModes::LeadingZeroMode::Print;
+ } else {
+ // LZ - processor-dependent (default behavior)
+ modes.leadingZero = MutableModes::LeadingZeroMode::Processor;
+ }
+ return;
+ }
+ break;
case 'S':
if (next == 'P') {
modes.editingFlags |= signPlus;
@@ -455,6 +470,7 @@ RT_API_ATTRS int FormatControl<CONTEXT>::CueUpNextDataEdit(
} else if (ch >= 'A' && ch <= 'Z') {
int start{offset_ - 1};
CharType next{'\0'};
+ CharType next2{'\0'};
if (ch != 'P') { // 1PE5.2 - comma not required (C1302)
CharType peek{Capitalize(PeekNext())};
if (peek >= 'A' && peek <= 'Z') {
@@ -464,6 +480,15 @@ RT_API_ATTRS int FormatControl<CONTEXT>::CueUpNextDataEdit(
// Assume a two-letter edit descriptor
next = peek;
++offset_;
+ } else if (ch == 'L' && peek == 'Z') {
+ // LZ, LZS, or LZP control edit descriptor
+ next = peek;
+ ++offset_;
+ CharType peek2{Capitalize(PeekNext())};
+ if (peek2 == 'S' || peek2 == 'P') {
+ next2 = peek2;
+ ++offset_;
+ }
} else {
// extension: assume a comma between 'ch' and 'peek'
}
@@ -484,7 +509,7 @@ RT_API_ATTRS int FormatControl<CONTEXT>::CueUpNextDataEdit(
repeat = GetIntField(context);
}
HandleControl(context, static_cast<char>(ch), static_cast<char>(next),
- repeat ? *repeat : 1);
+ repeat ? *repeat : 1, static_cast<char>(next2));
}
} else if (ch == '/') {
context.AdvanceRecord(repeat && *repeat > 0 ? *repeat : 1);
diff --git a/flang-rt/include/flang-rt/runtime/format.h b/flang-rt/include/flang-rt/runtime/format.h
index 79a7dd713b1a1..787772935ce8c 100644
--- a/flang-rt/include/flang-rt/runtime/format.h
+++ b/flang-rt/include/flang-rt/runtime/format.h
@@ -44,6 +44,12 @@ struct MutableModes {
return editingFlags & decimalComma ? char32_t{','} : char32_t{'.'};
}
+ enum class LeadingZeroMode : std::uint8_t {
+ Processor, // LZ: processor-dependent (default)
+ Suppress, // LZS: suppress optional leading zero
+ Print, // LZP: print optional leading zero
+ };
+
std::uint8_t editingFlags{0}; // BN, DP, SS
enum decimal::FortranRounding round{
executionEnvironment
@@ -53,6 +59,7 @@ struct MutableModes {
short scale{0}; // kP
bool inNamelist{false}; // skip ! comments
bool nonAdvancing{false}; // ADVANCE='NO', or $ or \ in FORMAT
+ LeadingZeroMode leadingZero{LeadingZeroMode::Processor}; // LZ/LZS/LZP
};
// A single edit descriptor extracted from a FORMAT
diff --git a/flang-rt/lib/runtime/edit-output.cpp b/flang-rt/lib/runtime/edit-output.cpp
index 73dba35ff08d9..d0aa8638102a9 100644
--- a/flang-rt/lib/runtime/edit-output.cpp
+++ b/flang-rt/lib/runtime/edit-output.cpp
@@ -411,10 +411,26 @@ RT_API_ATTRS bool RealOutputEditing<KIND>::EditEorDOutput(
if (totalLength > width || !exponent) {
return EmitRepeated(io_, '*', width);
}
- if (totalLength < width && digitsBeforePoint == 0 &&
- zeroesBeforePoint == 0) {
- zeroesBeforePoint = 1;
- ++totalLength;
+ if (digitsBeforePoint == 0 && zeroesBeforePoint == 0 && scale <= 0) {
+ // Optional leading zero position (F202X leading zero control).
+ // When scale > 0 (kP with k > 0), digits are moved before the decimal
+ // point, so the leading zero position is not optional -- skip this.
+ // Value has no digits before the decimal point: "0.xxxE+yy" vs ".xxxE+yy"
+ switch (edit.modes.leadingZero) {
+ case MutableModes::LeadingZeroMode::Print:
+ // LZP: always print the optional leading zero
+ zeroesBeforePoint = 1;
+ ++totalLength;
+ break;
+ case MutableModes::LeadingZeroMode::Suppress:
+ // LZS: never print the optional leading zero
+ break;
+ case MutableModes::LeadingZeroMode::Processor:
+ // LZ: processor-defined; flang chooses to print it
+ zeroesBeforePoint = 1;
+ ++totalLength;
+ break;
+ }
}
if (totalLength < width && editWidth == 0) {
width = totalLength;
@@ -544,7 +560,24 @@ RT_API_ATTRS bool RealOutputEditing<KIND>::EditFOutput(const DataEdit &edit) {
if (digitsBeforePoint + zeroesBeforePoint + zeroesAfterPoint +
digitsAfterPoint + trailingZeroes ==
0) {
- zeroesBeforePoint = 1; // "." -> "0."
+ zeroesBeforePoint = 1; // "."--> "0." (bare decimal point)
+ } else if (digitsBeforePoint == 0 && zeroesBeforePoint == 0 &&
+ expo <= 0) {
+ // Optional leading zero position (F202X leading zero control).
+ // Value magnitude < 1: "0.xxx" vs ".xxx"
+ switch (edit.modes.leadingZero) {
+ case MutableModes::LeadingZeroMode::Print:
+ // LZP: always print the optional leading zero
+ zeroesBeforePoint = 1;
+ break;
+ case MutableModes::LeadingZeroMode::Suppress:
+ // LZS: never print the optional leading zero
+ break;
+ case MutableModes::LeadingZeroMode::Processor:
+ // LZ: processor-defined; flang chooses to print it
+ zeroesBeforePoint = 1;
+ break;
+ }
}
int totalLength{signLength + digitsBeforePoint + zeroesBeforePoint +
1 /*'.'*/ + zeroesAfterPoint + digitsAfterPoint + trailingZeroes +
@@ -553,10 +586,6 @@ RT_API_ATTRS bool RealOutputEditing<KIND>::EditFOutput(const DataEdit &edit) {
if (totalLength > width) {
return EmitRepeated(io_, '*', width);
}
- if (totalLength < width && digitsBeforePoint + zeroesBeforePoint == 0) {
- zeroesBeforePoint = 1;
- ++totalLength;
- }
return EmitPrefix(edit, totalLength, width) &&
EmitAscii(io_, convertedStr, signLength + digitsBeforePoint) &&
EmitRepeated(io_, '0', zeroesBeforePoint) &&
diff --git a/flang/docs/F202X.md b/flang/docs/F202X.md
index d1940a1858db1..6a303981cf264 100644
--- a/flang/docs/F202X.md
+++ b/flang/docs/F202X.md
@@ -261,6 +261,15 @@ The `AT` edit descriptor automatically trims character output. The `LZP`,
`LZS`, and `LZ` control edit descriptors and `LEADING_ZERO=` specifier provide a
means for controlling the output of leading zero digits.
+Implementation status:
+- `LZ`, `LZS`, `LZP` control edit descriptors, affect only F, E, D, and G
+ editing of an output statement: Implemented
+ - `LZ` - Processor-dependent, default (e.g., `0.2`)
+ - `LZS` - Suppress leading zeros (e.g., `.2`)
+ - `LZP` - Print leading zero (e.g. `0.2`)
+- `AT` edit descriptor: Not yet implemented
+- `LEADING_ZERO=specifier` in OPEN statement: Not yet implemented
+
#### Intrinsic Module Extensions
Addressing some issues and omissions in intrinsic modules:
diff --git a/flang/docs/FortranStandardsSupport.md b/flang/docs/FortranStandardsSupport.md
index f57956cd6d6b8..02e4653280f12 100644
--- a/flang/docs/FortranStandardsSupport.md
+++ b/flang/docs/FortranStandardsSupport.md
@@ -48,7 +48,7 @@ status of all important Fortran 2023 features. The table entries are based on th
| Extensions for c_f_pointer intrinsic | Y | |
| Procedures for converting between fortran and c strings | N | |
| The at edit descriptor | N | |
-| Control over leading zeros in output of real values | N | |
+| Control over leading zeros in output of real values | P | LZ/LZS/LZP edit descriptors implemented; LEADING_ZERO=specifier not yet implemented |
| Extensions for Namelist | N | |
| Allow an object of a type with a coarray ultimate component to be an array or allocatable | N | |
| Put with Notify | N | |
diff --git a/flang/include/flang/Common/format.h b/flang/include/flang/Common/format.h
index 1e64acb823616..60bcc8e81ee38 100644
--- a/flang/include/flang/Common/format.h
+++ b/flang/include/flang/Common/format.h
@@ -114,7 +114,7 @@ struct FormatMessage {
// This declaration is logically private to class FormatValidator.
// It is placed here to work around a clang compilation problem.
ENUM_CLASS(TokenKind, None, A, B, BN, BZ, D, DC, DP, DT, E, EN, ES, EX, F, G, I,
- L, O, P, RC, RD, RN, RP, RU, RZ, S, SP, SS, T, TL, TR, X, Z, Colon, Slash,
+ L, LZ, LZP, LZS, O, P, RC, RD, RN, RP, RU, RZ, S, SP, SS, T, TL, TR, X, Z, Colon, Slash,
Backslash, // nonstandard: inhibit newline on output
Dollar, // nonstandard: inhibit newline on output on terminals
Star, LParen, RParen, Comma, Point, Sign,
@@ -219,7 +219,7 @@ template <typename CHAR = char> class FormatValidator {
std::int64_t knrValue_{-1}; // -1 ==> not present
std::int64_t scaleFactorValue_{}; // signed k in kP
std::int64_t wValue_{-1};
- char argString_[3]{}; // 1-2 character msg arg; usually edit descriptor name
+ char argString_[4]{}; // 1-3 character msg arg; usually edit descriptor name
bool formatHasErrors_{false};
bool unterminatedFormatError_{false};
bool suppressMessageCascade_{false};
@@ -390,7 +390,25 @@ template <typename CHAR> void FormatValidator<CHAR>::NextToken() {
token_.set_kind(TokenKind::I);
break;
case 'L':
- token_.set_kind(TokenKind::L);
+ switch (LookAheadChar()) {
+ case 'Z':
+ // Advance past 'Z', then look ahead for 'S' or 'P'
+ Advance(TokenKind::LZ);
+ switch (LookAheadChar()) {
+ case 'S':
+ Advance(TokenKind::LZS);
+ break;
+ case 'P':
+ Advance(TokenKind::LZP);
+ break;
+ default:
+ break;
+ }
+ break;
+ default:
+ token_.set_kind(TokenKind::L);
+ break;
+ }
break;
case 'O':
token_.set_kind(TokenKind::O);
@@ -674,9 +692,22 @@ template <typename CHAR> bool FormatValidator<CHAR>::Check() {
ReportError("Unexpected '%s' in format expression", signToken);
}
// Default message argument.
- // Alphabetic edit descriptor names are one or two characters in length.
+ // Alphabetic edit descriptor names are one to three characters in length.
argString_[0] = toupper(format_[token_.offset()]);
- argString_[1] = token_.length() > 1 ? toupper(*cursor_) : 0;
+ if (token_.length() > 2) {
+ // Three-character descriptor names (e.g., LZP, LZS).
+ // token_.offset() has the first character and *cursor_ has the last;
+ // find the middle character by scanning past any blanks.
+ const CHAR *mid{format_ + token_.offset() + 1};
+ while (mid < cursor_ && IsWhite(*mid)) {
+ ++mid;
+ }
+ argString_[1] = toupper(*mid);
+ argString_[2] = toupper(*cursor_);
+ } else {
+ argString_[1] = token_.length() > 1 ? toupper(*cursor_) : 0;
+ argString_[2] = 0;
+ }
// Process one format edit descriptor or do format list management.
switch (token_.kind()) {
case TokenKind::A:
@@ -794,6 +825,9 @@ template <typename CHAR> bool FormatValidator<CHAR>::Check() {
case TokenKind::BZ:
case TokenKind::DC:
case TokenKind::DP:
+ case TokenKind::LZ:
+ case TokenKind::LZS:
+ case TokenKind::LZP:
case TokenKind::RC:
case TokenKind::RD:
case TokenKind::RN:
@@ -807,6 +841,7 @@ template <typename CHAR> bool FormatValidator<CHAR>::Check() {
// R1318 blank-interp-edit-desc -> BN | BZ
// R1319 round-edit-desc -> RU | RD | RZ | RN | RC | RP
// R1320 decimal-edit-desc -> DC | DP
+ // F202X leading-zero-edit-desc -> LZ | LZS | LZP
check_r(false);
NextToken();
break;
diff --git a/flang/include/flang/Parser/format-specification.h b/flang/include/flang/Parser/format-specification.h
index 28c8affd7bde0..5d37a9c2c0060 100644
--- a/flang/include/flang/Parser/format-specification.h
+++ b/flang/include/flang/Parser/format-specification.h
@@ -95,6 +95,9 @@ struct ControlEditDesc {
RP,
DC,
DP,
+ LZ, // F202X: processor-dependent leading zero, default
+ LZS, // F202X: suppress leading zeros
+ LZP, // F202X: print leading zero
Dollar, // extension: inhibit newline on output
Backslash, // ditto, but only on terminals
};
diff --git a/flang/lib/Parser/io-parsers.cpp b/flang/lib/Parser/io-parsers.cpp
index cb3e68a05c94d..9da8c4f01d7dc 100644
--- a/flang/lib/Parser/io-parsers.cpp
+++ b/flang/lib/Parser/io-parsers.cpp
@@ -629,7 +629,8 @@ TYPE_PARSER(construct<format::IntrinsicTypeDataEditDesc>(
"X " >> pure(format::IntrinsicTypeDataEditDesc::Kind::EX) ||
pure(format::IntrinsicTypeDataEditDesc::Kind::E)) ||
"G " >> pure(format::IntrinsicTypeDataEditDesc::Kind::G) ||
- "L " >> pure(format::IntrinsicTypeDataEditDesc::Kind::L),
+ ("L "_tok / !letter /* don't occlude LZ, LZS, & LZP */) >>
+ pure(format::IntrinsicTypeDataEditDesc::Kind::L),
noInt, noInt, noInt)))
// R1307 data-edit-desc (part 2 of 2)
@@ -677,6 +678,12 @@ TYPE_PARSER(construct<format::ControlEditDesc>(
pure(format::ControlEditDesc::Kind::BN)) ||
"Z " >> construct<format::ControlEditDesc>(
pure(format::ControlEditDesc::Kind::BZ))) ||
+ "L " >> ("Z " >> ("S " >> construct<format::ControlEditDesc>(
+ pure(format::ControlEditDesc::Kind::LZS)) ||
+ "P " >> construct<format::ControlEditDesc>(
+ pure(format::ControlEditDesc::Kind::LZP)) ||
+ construct<format::ControlEditDesc>(
+ pure(format::ControlEditDesc::Kind::LZ)))) ||
"R " >> ("U " >> construct<format::ControlEditDesc>(
pure(format::ControlEditDesc::Kind::RU)) ||
"D " >> construct<format::ControlEditDesc>(
diff --git a/flang/lib/Parser/unparse.cpp b/flang/lib/Parser/unparse.cpp
index 3d8ea9f703b2f..8ad1f9b8ff618 100644
--- a/flang/lib/Parser/unparse.cpp
+++ b/flang/lib/Parser/unparse.cpp
@@ -1549,6 +1549,9 @@ class UnparseVisitor {
FMT(RP);
FMT(DC);
FMT(DP);
+ FMT(LZ);
+ FMT(LZS);
+ FMT(LZP);
#undef FMT
case format::ControlEditDesc::Kind::Dollar:
Put('$');
diff --git a/flang/test/Semantics/io16.f90 b/flang/test/Semantics/io16.f90
new file mode 100644
index 0000000000000..559d98f0d5ceb
--- /dev/null
+++ b/flang/test/Semantics/io16.f90
@@ -0,0 +1,99 @@
+! RUN: %python %S/test_errors.py %s %flang_fc1
+
+! F202X leading-zero control edit descriptors: LZ, LZS, LZP
+
+ ! Valid uses of LZ, LZP, LZS in FORMAT statements
+1001 format(LZ, F10.3)
+1002 format(LZP, F10.3)
+1003 format(LZS, F10.3)
+1004 format(LZ, E10.3)
+1005 format(LZP, E10.3)
+1006 format(LZS, E10.3)
+1007 format(LZS, D10.3)
+1008 format(LZ, G10.3)
+
+ ! Valid uses with blanks inside keywords (Fortran ignores blanks)
+1009 format(L Z, F10.3)
+1010 format(L Z P, F10.3)
+1011 format(L Z S, F10.3)
+
+ ! Combining with other control edit descriptors
+1012 format(LZP, DC, F10.3)
+1013 format(BN, LZS, F10.3)
+1014 format(LZ, SS, RZ, F10.3)
+
+ ! Multiple groups
+1015 format(LZP, 3F10.3, LZS, 2E12.4)
+
+ ! C1302 : multiple edit descriptors without ',' separation; no errors
+1016 format(LZF10.3)
+1017 format(LZPF10.3)
+1018 format(LZSF10.3)
+1019 format(LZE10.3)
+1020 format(LZPE10.3)
+1021 format(LZSD10.3)
+1022 format(LZG10.3)
+1023 format(LZPDCF10.3)
+1024 format(BNLZSF10.3)
+1025 format(LZPF10.3LZSF10.3)
+1026 format(LZP3F10.3LZS2E12.4)
+
+ ! In WRITE format strings
+ write(*, '(LZ, F10.3)') 0.5
+ write(*, '(LZP, F10.3)') 0.5
+ write(*, '(LZS, F10.3)') 0.5
+ write(*, '(LZP,E10.3)') 0.5
+ write(*, '(LZS,D10.3)') 0.5
+
+ ! C1302 : WRITE format strings without ',' separation; no errors
+ write(*, '(LZF10.3)') 0.5
+ write(*, '(LZPF10.3)') 0.5
+ write(*, '(LZSF10.3)') 0.5
+ write(*, '(LZPE10.3)') 0.5
+ write(*, '(LZP3F10.3LZS2E12.4)') 0.5, 0.5, 0.5, 0.5, 0.5
+
+ ! FMT= specifier with comma-separated descriptors
+ write(*, fmt='(LZ, F10.3)') 0.5
+ write(*, fmt='(LZP, F10.3)') 0.5
+ write(*, fmt='(LZS, F10.3)') 0.5
+ write(*, fmt='(LZP, E10.3)') 0.5
+ write(*, fmt='(LZS, D10.3)') 0.5
+ write(*, fmt='(LZP, DC, F10.3)') 0.5
+ write(*, fmt='(BN, LZS, F10.3)') 0.5
+
+ ! FMT= specifier without ',' separation; no errors
+ write(*, fmt='(LZF10.3)') 0.5
+ write(*, fmt='(LZPF10.3)') 0.5
+ write(*, fmt='(LZSF10.3)') 0.5
+ write(*, fmt='(LZPE10.3)') 0.5
+ write(*, fmt='(LZP3F10.3LZS2E12.4)') 0.5, 0.5, 0.5, 0.5, 0.5
+
+ ! FMT= specifier with FORMAT label reference
+ write(*, fmt=1001) 0.5
+ write(*, fmt=1002) 0.5
+ write(*, fmt=1017) 0.5
+
+ ! LZ/LZP/LZS coexisting with abbreviated L (no width) data edit descriptor
+ write(*, '(LZP, F10.3, L)') 0.5, .true.
+ write(*, '(LZS, F10.3, L)') 0.5, .true.
+
+ ! Error: repeat specifier before LZ/LZP/LZS in WRITE format strings
+ !ERROR: Repeat specifier before 'LZ' edit descriptor
+ write(*, '(3LZ, F10.3)') 0.5
+
+ !ERROR: Repeat specifier before 'LZP' edit descriptor
+ write(*, '(2LZP, F10.3)') 0.5
+
+ !ERROR: Repeat specifier before 'LZS' edit descriptor
+ write(*, '(2LZS, F10.3)') 0.5
+
+ ! Error: repeat specifier before LZ/LZP/LZS in FORMAT statements
+ !ERROR: Repeat specifier before 'LZ' edit descriptor
+2001 format(3LZ, F10.3)
+
+ !ERROR: Repeat specifier before 'LZP' edit descriptor
+2002 format(2LZP, F10.3)
+
+ !ERROR: Repeat specifier before 'LZS' edit descriptor
+2003 format(2LZS, F10.3)
+end
More information about the flang-commits
mailing list