[PATCH 0/3] Floating-point literal syntax extension for Ada

[PATCH 0/3] Floating-point literal syntax extension for Ada

[Tom Tromey]

This short series fixes a couple of small bugs, and adds a syntax
extension for floating-point literals for Ada.  The idea behind this
extension is that it is sometimes convenient to be able to specify a
floats underlying bits.

This needs a documentation review, but otherwise is completely
Ada-specific.

Tom

[PATCH 1/3] Fix Ada integer literals with exponents

[Tom Tromey]

While working on another patch, I noticed that Ada integer literals
with exponents did not work.  For example, with one form you get an
error:

    (gdb) print 8e2
    Invalid digit `e' in based literal

And with another form you get an incorrect value:

    (gdb) print 16#8#e2
    $2 = 8

This patch fixes the bugs and adds tests.
---
 gdb/ada-lex.l                      |  9 +++++---
 gdb/testsuite/gdb.ada/literals.exp | 36 ++++++++++++++++++++++++++++++
 2 files changed, 42 insertions(+), 3 deletions(-)
 create mode 100644 gdb/testsuite/gdb.ada/literals.exp

diff --git a/gdb/ada-lex.l b/gdb/ada-lex.l
index f61efba81a9..aab53b397ec 100644
--- a/gdb/ada-lex.l
+++ b/gdb/ada-lex.l
@@ -103,8 +103,9 @@ static int paren_depth;
 
 {NUM10}{POSEXP}  {
 		   canonicalizeNumeral (numbuf, yytext);
-		   return processInt (pstate, NULL, numbuf,
-				      strrchr (numbuf, 'e') + 1);
+		   char *e_ptr = strrchr (numbuf, 'e');
+		   *e_ptr = '\0';
+		   return processInt (pstate, nullptr, numbuf, e_ptr + 1);
 		 }
 
 {NUM10}          {
@@ -114,9 +115,11 @@ static int paren_depth;
 
 {NUM10}"#"{HEXDIG}({HEXDIG}|_)*"#"{POSEXP} {
 		   canonicalizeNumeral (numbuf, yytext);
+		   char *e_ptr = strrchr (numbuf, 'e');
+		   *e_ptr = '\0';
 		   return processInt (pstate, numbuf,
 				      strchr (numbuf, '#') + 1,
-				      strrchr(numbuf, '#') + 1);
+				      e_ptr + 1);
 		 }
 
 {NUM10}"#"{HEXDIG}({HEXDIG}|_)*"#" {
diff --git a/gdb/testsuite/gdb.ada/literals.exp b/gdb/testsuite/gdb.ada/literals.exp
new file mode 100644
index 00000000000..92a9a1954fc
--- /dev/null
+++ b/gdb/testsuite/gdb.ada/literals.exp
@@ -0,0 +1,36 @@
+# Copyright 2022 Free Software Foundation, Inc.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# Test some literal syntax.
+
+load_lib "ada.exp"
+
+if { [skip_ada_tests] } { return -1 }
+
+clean_restart
+
+gdb_test_no_output "set lang ada"
+gdb_test "print 7#10#" " = 7"
+gdb_test "print 77#10#" "Invalid base: 77."
+gdb_test "print 7#8#" "Invalid digit `8' in based literal"
+
+gdb_test "print 8e2" " = 800"
+gdb_test "print 9999999999999999999999999999999999999999999999" \
+    "Integer literal out of range"
+gdb_test "print 2e1000" "Integer literal out of range"
+
+gdb_test "print 16#ffff#" " = 65535"
+gdb_test "print 16#f#e1" " = 240"
+gdb_test "print 16#1#e10" " = 1099511627776"
-- 
2.31.1

[PATCH 2/3] Implement real literal extension for Ada

[Tom Tromey]

Sometimes it is convenient to be able to specify the exact bits of a
floating-point literal.  For example, you may want to set a
floating-point register to a denormalized value, or to a particular
NaN.

In C, you can do this by combining the "{}" cast with an array
literal, like:

    (gdb) p {double}{0x576488BDD2AE9FFE}
    $1 = 9.8765449999999996e+112

This patch adds a somewhat similar idea to Ada.  It extends the lexer
to allow "l" and "f" suffixes in a based literal.  The "f" indicates a
floating-point literal, and the "l"s control the size of the
floating-point type.

Note that this differs from Ada's based real literals.  I believe
those can also be used to control the bits of a floating-point value,
but they are a bit more cumbersome to use (simplest is binary but
that's also very lengthy).  Also, these aren't implemented in GDB.

I chose not to allow this extension to work with based integer
literals with exponents.  That didn't seem very useful.
---
 gdb/NEWS                                  |  8 ++
 gdb/ada-lex.l                             | 98 ++++++++++++++++++-----
 gdb/doc/gdb.texinfo                       | 16 ++++
 gdb/testsuite/gdb.ada/float-bits.exp      | 50 ++++++++++++
 gdb/testsuite/gdb.ada/float-bits/prog.adb | 22 +++++
 5 files changed, 174 insertions(+), 20 deletions(-)
 create mode 100644 gdb/testsuite/gdb.ada/float-bits.exp
 create mode 100644 gdb/testsuite/gdb.ada/float-bits/prog.adb

diff --git a/gdb/NEWS b/gdb/NEWS
index dc2cac1871b..2876ca822ad 100644
--- a/gdb/NEWS
+++ b/gdb/NEWS
@@ -136,6 +136,14 @@ info win
   This command now includes information about the width of the tui
   windows in its output.
 
+* GDB's Ada parser now supports an extension for specifying the exact
+  byte contents of a floating-point literal.  This can be useful for
+  setting floating-point registers to a precise value without loss of
+  precision.  The syntax is an extension of the based literal syntax.
+  Use, e.g., "16lf#0123abcd#" -- the number of "l"s controls the width
+  of the floating-point type, and the "f" is the marker for floating
+  point.
+
 * New targets
 
 GNU/Linux/LoongArch    loongarch*-*-linux*
diff --git a/gdb/ada-lex.l b/gdb/ada-lex.l
index aab53b397ec..8e64d3a2926 100644
--- a/gdb/ada-lex.l
+++ b/gdb/ada-lex.l
@@ -122,7 +122,11 @@ static int paren_depth;
 				      e_ptr + 1);
 		 }
 
-{NUM10}"#"{HEXDIG}({HEXDIG}|_)*"#" {
+	/* The "llf" is a gdb extension to allow a floating-point
+	   constant to be written in some other base.  The
+	   floating-point number is formed by reinterpreting the
+	   bytes, allowing direct control over the bits.  */
+{NUM10}(l{0,2}f)?"#"{HEXDIG}({HEXDIG}|_)*"#" {
 		   canonicalizeNumeral (numbuf, yytext);
 		   return processInt (pstate, numbuf, strchr (numbuf, '#') + 1,
 				      NULL);
@@ -347,18 +351,36 @@ static int
 processInt (struct parser_state *par_state, const char *base0,
 	    const char *num0, const char *exp0)
 {
-  ULONGEST result;
   long exp;
   int base;
-  const char *trailer;
+  /* For the based literal with an "f" prefix, we'll return a
+     floating-point number.  This counts the the number of "l"s seen,
+     to decide the width of the floating-point number to return.  -1
+     means no "f".  */
+  int floating_point_l_count = -1;
 
   if (base0 == NULL)
     base = 10;
   else
     {
-      base = strtol (base0, (char **) NULL, 10);
+      char *end_of_base;
+      base = strtol (base0, &end_of_base, 10);
       if (base < 2 || base > 16)
 	error (_("Invalid base: %d."), base);
+      while (*end_of_base == 'l')
+	{
+	  ++floating_point_l_count;
+	  ++end_of_base;
+	}
+      /* This assertion is ensured by the pattern.  */
+      gdb_assert (floating_point_l_count == -1 || *end_of_base == 'f');
+      if (*end_of_base == 'f')
+	{
+	  ++end_of_base;
+	  ++floating_point_l_count;
+	}
+      /* This assertion is ensured by the pattern.  */
+      gdb_assert (*end_of_base == '#');
     }
 
   if (exp0 == NULL)
@@ -366,26 +388,62 @@ processInt (struct parser_state *par_state, const char *base0,
   else
     exp = strtol(exp0, (char **) NULL, 10);
 
-  errno = 0;
-  result = strtoulst (num0, &trailer, base);
-  if (errno == ERANGE)
-    error (_("Integer literal out of range"));
-  if (isxdigit(*trailer))
-    error (_("Invalid digit `%c' in based literal"), *trailer);
+  gdb_mpz result;
+  while (isxdigit (*num0))
+    {
+      int dig = fromhex (*num0);
+      if (dig >= base)
+	error (_("Invalid digit `%c' in based literal"), *num0);
+      mpz_mul_ui (result.val, result.val, base);
+      mpz_add_ui (result.val, result.val, dig);
+      ++num0;
+    }
 
   while (exp > 0)
     {
-      if (result > (ULONG_MAX / base))
-	error (_("Integer literal out of range"));
-      result *= base;
+      mpz_mul_ui (result.val, result.val, base);
       exp -= 1;
     }
 
-  if ((result >> (gdbarch_int_bit (par_state->gdbarch ())-1)) == 0)
+  if (floating_point_l_count > -1)
+    {
+      struct type *fp_type;
+      if (floating_point_l_count == 0)
+	fp_type = language_lookup_primitive_type (par_state->language (),
+						  par_state->gdbarch (),
+						  "float");
+      else if (floating_point_l_count == 1)
+	fp_type = language_lookup_primitive_type (par_state->language (),
+						  par_state->gdbarch (),
+						  "long_float");
+      else
+	{
+	  /* This assertion is ensured by the pattern.  */
+	  gdb_assert (floating_point_l_count == 2);
+	  fp_type = language_lookup_primitive_type (par_state->language (),
+						    par_state->gdbarch (),
+						    "long_long_float");
+	}
+
+      yylval.typed_val_float.type = fp_type;
+      result.write (gdb::make_array_view (yylval.typed_val_float.val,
+					  TYPE_LENGTH (fp_type)),
+		    type_byte_order (fp_type),
+		    true);
+
+      return FLOAT;
+    }
+
+  gdb_mpz maxval (ULONGEST_MAX / base);
+  if (mpz_cmp (result.val, maxval.val) > 0)
+    error (_("Integer literal out of range"));
+
+  LONGEST value = result.as_integer<LONGEST> ();
+  if ((value >> (gdbarch_int_bit (par_state->gdbarch ())-1)) == 0)
     yylval.typed_val.type = type_int (par_state);
-  else if ((result >> (gdbarch_long_bit (par_state->gdbarch ())-1)) == 0)
+  else if ((value >> (gdbarch_long_bit (par_state->gdbarch ())-1)) == 0)
     yylval.typed_val.type = type_long (par_state);
-  else if (((result >> (gdbarch_long_bit (par_state->gdbarch ())-1)) >> 1) == 0)
+  else if (((value >> (gdbarch_long_bit (par_state->gdbarch ())-1)) >> 1) == 0)
     {
       /* We have a number representable as an unsigned integer quantity.
          For consistency with the C treatment, we will treat it as an
@@ -396,18 +454,18 @@ processInt (struct parser_state *par_state, const char *base0,
        */
       yylval.typed_val.type
 	= builtin_type (par_state->gdbarch ())->builtin_unsigned_long;
-      if (result & LONGEST_SIGN)
+      if (value & LONGEST_SIGN)
 	yylval.typed_val.val =
-	  (LONGEST) (result & ~LONGEST_SIGN)
+	  (LONGEST) (value & ~LONGEST_SIGN)
 	  - (LONGEST_SIGN>>1) - (LONGEST_SIGN>>1);
       else
-	yylval.typed_val.val = (LONGEST) result;
+	yylval.typed_val.val = (LONGEST) value;
       return INT;
     }
   else
     yylval.typed_val.type = type_long_long (par_state);
 
-  yylval.typed_val.val = (LONGEST) result;
+  yylval.typed_val.val = value;
   return INT;
 }
 
diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
index f7f5f7a6158..9d5f9b5aa94 100644
--- a/gdb/doc/gdb.texinfo
+++ b/gdb/doc/gdb.texinfo
@@ -18189,6 +18189,9 @@ context.
 Should your program
 redefine these names in a package or procedure (at best a dubious practice),
 you will have to use fully qualified names to access their new definitions.
+
+@item
+Based real literals are not implemented.
 @end itemize
 
 @node Additions to Ada
@@ -18247,6 +18250,19 @@ complex conditional breaks:
 (@value{GDBP}) condition 1 (report(i); k += 1; A(k) > 100)
 @end smallexample
 
+@item
+An extension to based literals can be used to specify the exact bits
+of a real value.  After the base, you can use from zero to two
+@samp{l} characters, followed by an @samp{f}.  The number of @samp{l}
+characters controls the width of the resulting real constant: zero
+means @code{Float} is used, one means @code{Long_Float}, and two means
+@code{Long_Long_Float}.
+
+@smallexample
+(@value{GDBP}) print 16f#41b80000#
+$1 = 23.0
+@end smallexample
+
 @item 
 Rather than use catenation and symbolic character names to introduce special 
 characters into strings, one may instead use a special bracket notation, 
diff --git a/gdb/testsuite/gdb.ada/float-bits.exp b/gdb/testsuite/gdb.ada/float-bits.exp
new file mode 100644
index 00000000000..61db5f325ad
--- /dev/null
+++ b/gdb/testsuite/gdb.ada/float-bits.exp
@@ -0,0 +1,50 @@
+# Copyright 2022 Free Software Foundation, Inc.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# Test floating-point literal extension.
+
+load_lib "ada.exp"
+
+if { [skip_ada_tests] } { return -1 }
+
+standard_ada_testfile prog
+
+if {[gdb_compile_ada "${srcfile}" "${binfile}" executable {debug}] != ""} {
+    return -1
+}
+
+clean_restart ${testfile}
+
+set bp_location [gdb_get_line_number "BREAK" ${testdir}/prog.adb]
+runto "prog.adb:$bp_location"
+
+gdb_test "print 16f#41b80000#" " = 23.0"
+gdb_test "print val_float" " = 23.0"
+gdb_test "print val_float := 16f#41b80000#" " = 23.0"
+gdb_test "print val_float" " = 23.0" \
+    "print val_float after assignment"
+
+gdb_test "print 16lf#bc0d83c94fb6d2ac#" " = -2.0e-19"
+gdb_test "print val_double" " = -2.0e-19"
+gdb_test "print val_double := 16lf#bc0d83c94fb6d2ac#" " = -2.0e-19"
+gdb_test "print val_double" " = -2.0e-19" \
+    "print val_double after assignment"
+
+gdb_test "print 16llf#7FFFF7FF4054A56FA5B99019A5C8#" " = 5.0e\\+25"
+gdb_test "print val_long_double" " = 5.0e\\+25"
+gdb_test "print val_long_double := 16llf#7FFFF7FF4054A56FA5B99019A5C8#" \
+    " = 5.0e\\+25"
+gdb_test "print val_long_double" " = 5.0e\\+25" \
+    "print val_long_double after assignment"
diff --git a/gdb/testsuite/gdb.ada/float-bits/prog.adb b/gdb/testsuite/gdb.ada/float-bits/prog.adb
new file mode 100644
index 00000000000..0d8c18f8d47
--- /dev/null
+++ b/gdb/testsuite/gdb.ada/float-bits/prog.adb
@@ -0,0 +1,22 @@
+--  Copyright 2022 Free Software Foundation, Inc.
+--
+--  This program is free software; you can redistribute it and/or modify
+--  it under the terms of the GNU General Public License as published by
+--  the Free Software Foundation; either version 3 of the License, or
+--  (at your option) any later version.
+--
+--  This program is distributed in the hope that it will be useful,
+--  but WITHOUT ANY WARRANTY; without even the implied warranty of
+--  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+--  GNU General Public License for more details.
+--
+--  You should have received a copy of the GNU General Public License
+--  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+procedure Prog is
+   Val_Float : Float := 23.0;
+   Val_Double : Long_Float := -2.0e-19;
+   Val_Long_Double : Long_Long_Float := 5.0e+25;
+begin
+   null; -- BREAK
+end Prog;
-- 
2.31.1

[PATCH 3/3] Fix bug in ada_print_floating

[Tom Tromey]

ada_print_floating rewrites a floating-point string representation to
conform to Ada syntax.  However, if you managed to get a floating
point error, you might see:

    (gdb) print whatever
    $2 = <invalid float valu.0e>

What's happening here is that ada_print_floating doesn't recognize
this error case, and proceeds to modify the error text.

This patch fixes this problem.
---
 gdb/ada-valprint.c                   | 7 +++++++
 gdb/testsuite/gdb.ada/float-bits.exp | 3 +++
 2 files changed, 10 insertions(+)

diff --git a/gdb/ada-valprint.c b/gdb/ada-valprint.c
index bf95719f040..e113088491c 100644
--- a/gdb/ada-valprint.c
+++ b/gdb/ada-valprint.c
@@ -314,6 +314,13 @@ ada_print_floating (const gdb_byte *valaddr, struct type *type,
   std::string s = tmp_stream.release ();
   size_t skip_count = 0;
 
+  /* Don't try to modify a result representing an error.  */
+  if (s[0] == '<')
+    {
+      fputs_filtered (s.c_str (), stream);
+      return;
+    }
+
   /* Modify for Ada rules.  */
 
   size_t pos = s.find ("inf");
diff --git a/gdb/testsuite/gdb.ada/float-bits.exp b/gdb/testsuite/gdb.ada/float-bits.exp
index 61db5f325ad..c98afb53c06 100644
--- a/gdb/testsuite/gdb.ada/float-bits.exp
+++ b/gdb/testsuite/gdb.ada/float-bits.exp
@@ -48,3 +48,6 @@ gdb_test "print val_long_double := 16llf#7FFFF7FF4054A56FA5B99019A5C8#" \
     " = 5.0e\\+25"
 gdb_test "print val_long_double" " = 5.0e\\+25" \
     "print val_long_double after assignment"
+
+gdb_test "print 16llf#a56fa5b99019a5c800007ffff7ff4054#" \
+    " = <invalid float value>"
-- 
2.31.1

Re: [PATCH 2/3] Implement real literal extension for Ada

[Eli Zaretskii]

> Date: Tue,  1 Mar 2022 10:00:54 -0700
> From: Tom Tromey via Gdb-patches <gdb-patches@sourceware.org>
> Cc: Tom Tromey <tromey@adacore.com>
> 
> +* GDB's Ada parser now supports an extension for specifying the exact
> +  byte contents of a floating-point literal.  This can be useful for

"Exact byte contents" is not the best way to describe this
specification of a value.  If no better idea emerges, how about using
the same description as you used in the manual?

Otherwise, the documentation part is OK.  Thanks.

Re: [PATCH 2/3] Implement real literal extension for Ada

[Tom Tromey]

Eli> "Exact byte contents" is not the best way to describe this
Eli> specification of a value.  If no better idea emerges, how about using
Eli> the same description as you used in the manual?

I did this, thanks.

Tom

Re: [PATCH 0/3] Floating-point literal syntax extension for Ada

[Tom Tromey]

>>>>> "Tom" == Tom Tromey via Gdb-patches <gdb-patches@sourceware.org> writes:

Tom> This short series fixes a couple of small bugs, and adds a syntax
Tom> extension for floating-point literals for Ada.  The idea behind this
Tom> extension is that it is sometimes convenient to be able to specify a
Tom> floats underlying bits.

Tom> This needs a documentation review, but otherwise is completely
Tom> Ada-specific.

I'm checking this in now.

Tom

Re: [PATCH 2/3] Implement real literal extension for Ada

[Luis Machado]

Hi Tom,

On 3/1/22 17:00, Tom Tromey via Gdb-patches wrote:
> Sometimes it is convenient to be able to specify the exact bits of a
> floating-point literal.  For example, you may want to set a
> floating-point register to a denormalized value, or to a particular
> NaN.
> 
> In C, you can do this by combining the "{}" cast with an array
> literal, like:
> 
>      (gdb) p {double}{0x576488BDD2AE9FFE}
>      $1 = 9.8765449999999996e+112
> 
> This patch adds a somewhat similar idea to Ada.  It extends the lexer
> to allow "l" and "f" suffixes in a based literal.  The "f" indicates a
> floating-point literal, and the "l"s control the size of the
> floating-point type.
> 
> Note that this differs from Ada's based real literals.  I believe
> those can also be used to control the bits of a floating-point value,
> but they are a bit more cumbersome to use (simplest is binary but
> that's also very lengthy).  Also, these aren't implemented in GDB.
> 
> I chose not to allow this extension to work with based integer
> literals with exponents.  That didn't seem very useful.
> ---
>   gdb/NEWS                                  |  8 ++
>   gdb/ada-lex.l                             | 98 ++++++++++++++++++-----
>   gdb/doc/gdb.texinfo                       | 16 ++++
>   gdb/testsuite/gdb.ada/float-bits.exp      | 50 ++++++++++++
>   gdb/testsuite/gdb.ada/float-bits/prog.adb | 22 +++++
>   5 files changed, 174 insertions(+), 20 deletions(-)
>   create mode 100644 gdb/testsuite/gdb.ada/float-bits.exp
>   create mode 100644 gdb/testsuite/gdb.ada/float-bits/prog.adb
> 
> diff --git a/gdb/NEWS b/gdb/NEWS
> index dc2cac1871b..2876ca822ad 100644
> --- a/gdb/NEWS
> +++ b/gdb/NEWS
> @@ -136,6 +136,14 @@ info win
>     This command now includes information about the width of the tui
>     windows in its output.
>   
> +* GDB's Ada parser now supports an extension for specifying the exact
> +  byte contents of a floating-point literal.  This can be useful for
> +  setting floating-point registers to a precise value without loss of
> +  precision.  The syntax is an extension of the based literal syntax.
> +  Use, e.g., "16lf#0123abcd#" -- the number of "l"s controls the width
> +  of the floating-point type, and the "f" is the marker for floating
> +  point.
> +
>   * New targets
>   
>   GNU/Linux/LoongArch    loongarch*-*-linux*
> diff --git a/gdb/ada-lex.l b/gdb/ada-lex.l
> index aab53b397ec..8e64d3a2926 100644
> --- a/gdb/ada-lex.l
> +++ b/gdb/ada-lex.l
> @@ -122,7 +122,11 @@ static int paren_depth;
>   				      e_ptr + 1);
>   		 }
>   
> -{NUM10}"#"{HEXDIG}({HEXDIG}|_)*"#" {
> +	/* The "llf" is a gdb extension to allow a floating-point
> +	   constant to be written in some other base.  The
> +	   floating-point number is formed by reinterpreting the
> +	   bytes, allowing direct control over the bits.  */
> +{NUM10}(l{0,2}f)?"#"{HEXDIG}({HEXDIG}|_)*"#" {
>   		   canonicalizeNumeral (numbuf, yytext);
>   		   return processInt (pstate, numbuf, strchr (numbuf, '#') + 1,
>   				      NULL);
> @@ -347,18 +351,36 @@ static int
>   processInt (struct parser_state *par_state, const char *base0,
>   	    const char *num0, const char *exp0)
>   {
> -  ULONGEST result;
>     long exp;
>     int base;
> -  const char *trailer;
> +  /* For the based literal with an "f" prefix, we'll return a
> +     floating-point number.  This counts the the number of "l"s seen,
> +     to decide the width of the floating-point number to return.  -1
> +     means no "f".  */
> +  int floating_point_l_count = -1;
>   
>     if (base0 == NULL)
>       base = 10;
>     else
>       {
> -      base = strtol (base0, (char **) NULL, 10);
> +      char *end_of_base;
> +      base = strtol (base0, &end_of_base, 10);
>         if (base < 2 || base > 16)
>   	error (_("Invalid base: %d."), base);
> +      while (*end_of_base == 'l')
> +	{
> +	  ++floating_point_l_count;
> +	  ++end_of_base;
> +	}
> +      /* This assertion is ensured by the pattern.  */
> +      gdb_assert (floating_point_l_count == -1 || *end_of_base == 'f');
> +      if (*end_of_base == 'f')
> +	{
> +	  ++end_of_base;
> +	  ++floating_point_l_count;
> +	}
> +      /* This assertion is ensured by the pattern.  */
> +      gdb_assert (*end_of_base == '#');
>       }
>   
>     if (exp0 == NULL)
> @@ -366,26 +388,62 @@ processInt (struct parser_state *par_state, const char *base0,
>     else
>       exp = strtol(exp0, (char **) NULL, 10);
>   
> -  errno = 0;
> -  result = strtoulst (num0, &trailer, base);
> -  if (errno == ERANGE)
> -    error (_("Integer literal out of range"));
> -  if (isxdigit(*trailer))
> -    error (_("Invalid digit `%c' in based literal"), *trailer);
> +  gdb_mpz result;
> +  while (isxdigit (*num0))
> +    {
> +      int dig = fromhex (*num0);
> +      if (dig >= base)
> +	error (_("Invalid digit `%c' in based literal"), *num0);
> +      mpz_mul_ui (result.val, result.val, base);
> +      mpz_add_ui (result.val, result.val, dig);
> +      ++num0;
> +    }
>   
>     while (exp > 0)
>       {
> -      if (result > (ULONG_MAX / base))
> -	error (_("Integer literal out of range"));
> -      result *= base;
> +      mpz_mul_ui (result.val, result.val, base);
>         exp -= 1;
>       }
>   
> -  if ((result >> (gdbarch_int_bit (par_state->gdbarch ())-1)) == 0)
> +  if (floating_point_l_count > -1)
> +    {
> +      struct type *fp_type;
> +      if (floating_point_l_count == 0)
> +	fp_type = language_lookup_primitive_type (par_state->language (),
> +						  par_state->gdbarch (),
> +						  "float");
> +      else if (floating_point_l_count == 1)
> +	fp_type = language_lookup_primitive_type (par_state->language (),
> +						  par_state->gdbarch (),
> +						  "long_float");
> +      else
> +	{
> +	  /* This assertion is ensured by the pattern.  */
> +	  gdb_assert (floating_point_l_count == 2);
> +	  fp_type = language_lookup_primitive_type (par_state->language (),
> +						    par_state->gdbarch (),
> +						    "long_long_float");
> +	}
> +
> +      yylval.typed_val_float.type = fp_type;
> +      result.write (gdb::make_array_view (yylval.typed_val_float.val,
> +					  TYPE_LENGTH (fp_type)),
> +		    type_byte_order (fp_type),
> +		    true);
> +
> +      return FLOAT;
> +    }
> +
> +  gdb_mpz maxval (ULONGEST_MAX / base);
> +  if (mpz_cmp (result.val, maxval.val) > 0)
> +    error (_("Integer literal out of range"));
> +
> +  LONGEST value = result.as_integer<LONGEST> ();
> +  if ((value >> (gdbarch_int_bit (par_state->gdbarch ())-1)) == 0)
>       yylval.typed_val.type = type_int (par_state);
> -  else if ((result >> (gdbarch_long_bit (par_state->gdbarch ())-1)) == 0)
> +  else if ((value >> (gdbarch_long_bit (par_state->gdbarch ())-1)) == 0)
>       yylval.typed_val.type = type_long (par_state);
> -  else if (((result >> (gdbarch_long_bit (par_state->gdbarch ())-1)) >> 1) == 0)
> +  else if (((value >> (gdbarch_long_bit (par_state->gdbarch ())-1)) >> 1) == 0)
>       {
>         /* We have a number representable as an unsigned integer quantity.
>            For consistency with the C treatment, we will treat it as an
> @@ -396,18 +454,18 @@ processInt (struct parser_state *par_state, const char *base0,
>          */
>         yylval.typed_val.type
>   	= builtin_type (par_state->gdbarch ())->builtin_unsigned_long;
> -      if (result & LONGEST_SIGN)
> +      if (value & LONGEST_SIGN)
>   	yylval.typed_val.val =
> -	  (LONGEST) (result & ~LONGEST_SIGN)
> +	  (LONGEST) (value & ~LONGEST_SIGN)
>   	  - (LONGEST_SIGN>>1) - (LONGEST_SIGN>>1);
>         else
> -	yylval.typed_val.val = (LONGEST) result;
> +	yylval.typed_val.val = (LONGEST) value;
>         return INT;
>       }
>     else
>       yylval.typed_val.type = type_long_long (par_state);
>   
> -  yylval.typed_val.val = (LONGEST) result;
> +  yylval.typed_val.val = value;
>     return INT;
>   }
>   
> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
> index f7f5f7a6158..9d5f9b5aa94 100644
> --- a/gdb/doc/gdb.texinfo
> +++ b/gdb/doc/gdb.texinfo
> @@ -18189,6 +18189,9 @@ context.
>   Should your program
>   redefine these names in a package or procedure (at best a dubious practice),
>   you will have to use fully qualified names to access their new definitions.
> +
> +@item
> +Based real literals are not implemented.
>   @end itemize
>   
>   @node Additions to Ada
> @@ -18247,6 +18250,19 @@ complex conditional breaks:
>   (@value{GDBP}) condition 1 (report(i); k += 1; A(k) > 100)
>   @end smallexample
>   
> +@item
> +An extension to based literals can be used to specify the exact bits
> +of a real value.  After the base, you can use from zero to two
> +@samp{l} characters, followed by an @samp{f}.  The number of @samp{l}
> +characters controls the width of the resulting real constant: zero
> +means @code{Float} is used, one means @code{Long_Float}, and two means
> +@code{Long_Long_Float}.
> +
> +@smallexample
> +(@value{GDBP}) print 16f#41b80000#
> +$1 = 23.0
> +@end smallexample
> +
>   @item
>   Rather than use catenation and symbolic character names to introduce special
>   characters into strings, one may instead use a special bracket notation,
> diff --git a/gdb/testsuite/gdb.ada/float-bits.exp b/gdb/testsuite/gdb.ada/float-bits.exp
> new file mode 100644
> index 00000000000..61db5f325ad
> --- /dev/null
> +++ b/gdb/testsuite/gdb.ada/float-bits.exp
> @@ -0,0 +1,50 @@
> +# Copyright 2022 Free Software Foundation, Inc.
> +#
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
> +
> +# Test floating-point literal extension.
> +
> +load_lib "ada.exp"
> +
> +if { [skip_ada_tests] } { return -1 }
> +
> +standard_ada_testfile prog
> +
> +if {[gdb_compile_ada "${srcfile}" "${binfile}" executable {debug}] != ""} {
> +    return -1
> +}
> +
> +clean_restart ${testfile}
> +
> +set bp_location [gdb_get_line_number "BREAK" ${testdir}/prog.adb]
> +runto "prog.adb:$bp_location"
> +
> +gdb_test "print 16f#41b80000#" " = 23.0"
> +gdb_test "print val_float" " = 23.0"
> +gdb_test "print val_float := 16f#41b80000#" " = 23.0"
> +gdb_test "print val_float" " = 23.0" \
> +    "print val_float after assignment"
> +
> +gdb_test "print 16lf#bc0d83c94fb6d2ac#" " = -2.0e-19"
> +gdb_test "print val_double" " = -2.0e-19"
> +gdb_test "print val_double := 16lf#bc0d83c94fb6d2ac#" " = -2.0e-19"
> +gdb_test "print val_double" " = -2.0e-19" \
> +    "print val_double after assignment"
> +
> +gdb_test "print 16llf#7FFFF7FF4054A56FA5B99019A5C8#" " = 5.0e\\+25"
> +gdb_test "print val_long_double" " = 5.0e\\+25"
> +gdb_test "print val_long_double := 16llf#7FFFF7FF4054A56FA5B99019A5C8#" \
> +    " = 5.0e\\+25"
> +gdb_test "print val_long_double" " = 5.0e\\+25" \
> +    "print val_long_double after assignment"
> diff --git a/gdb/testsuite/gdb.ada/float-bits/prog.adb b/gdb/testsuite/gdb.ada/float-bits/prog.adb
> new file mode 100644
> index 00000000000..0d8c18f8d47
> --- /dev/null
> +++ b/gdb/testsuite/gdb.ada/float-bits/prog.adb
> @@ -0,0 +1,22 @@
> +--  Copyright 2022 Free Software Foundation, Inc.
> +--
> +--  This program is free software; you can redistribute it and/or modify
> +--  it under the terms of the GNU General Public License as published by
> +--  the Free Software Foundation; either version 3 of the License, or
> +--  (at your option) any later version.
> +--
> +--  This program is distributed in the hope that it will be useful,
> +--  but WITHOUT ANY WARRANTY; without even the implied warranty of
> +--  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +--  GNU General Public License for more details.
> +--
> +--  You should have received a copy of the GNU General Public License
> +--  along with this program.  If not, see <http://www.gnu.org/licenses/>.
> +
> +procedure Prog is
> +   Val_Float : Float := 23.0;
> +   Val_Double : Long_Float := -2.0e-19;
> +   Val_Long_Double : Long_Long_Float := 5.0e+25;
> +begin
> +   null; -- BREAK
> +end Prog;


I think the floating point representations may differ between 
architectures/bit-sizes.

I noticed 5 failures in this test for aarch64-linux and arm-linux.

FAIL: gdb.ada/float-bits.exp: print 16llf#7FFFF7FF4054A56FA5B99019A5C8#
FAIL: gdb.ada/float-bits.exp: print val_long_double
FAIL: gdb.ada/float-bits.exp: print val_long_double := 
16llf#7FFFF7FF4054A56FA5B99019A5C8#
FAIL: gdb.ada/float-bits.exp: print val_long_double after assignment
FAIL: gdb.ada/float-bits.exp: print

Some of the expected values don't really match. If the output is 
supposed to differ, maybe we should compare hex values instead?

Also, for 32-bit targets, I'm seeing size-related issues.

Here's one example for aarch64:

(expected)

(gdb) print 16llf#7FFFF7FF4054A56FA5B99019A5C8#^M
$9 = 5.0e+25

(actual)

(gdb) print 16llf#7FFFF7FF4054A56FA5B99019A5C8#
$9 = 1.68104996779424899119072698287782049e-4932

And for arm 32-bit;

(expected)

(gdb) print 16llf#7FFFF7FF4054A56FA5B99019A5C8#^M
$9 = 5.0e+25

(actual)

(gdb) print 16llf#7FFFF7FF4054A56FA5B99019A5C8#
Cannot export value 2596145952482202326224873165792712 as 64-bits 
unsigned integer (must be between 0 and 18446744073709551615)

Regards,
Luis

Re: [PATCH 2/3] Implement real literal extension for Ada

[Tom Tromey]

>>>>> "Luis" == Luis Machado <luis.machado@arm.com> writes:

Luis> I think the floating point representations may differ between
Luis> architectures/bit-sizes.

Luis> I noticed 5 failures in this test for aarch64-linux and arm-linux.

This is interesting, because internally we have a rewrite of this same
test, and it seems fine on aarch64-linux.

Luis> FAIL: gdb.ada/float-bits.exp: print 16llf#7FFFF7FF4054A56FA5B99019A5C8#
Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double
Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double :=
Luis> 16llf#7FFFF7FF4054A56FA5B99019A5C8#
Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double after assignment
Luis> FAIL: gdb.ada/float-bits.exp: print

Luis> Some of the expected values don't really match. If the output is
Luis> supposed to differ, maybe we should compare hex values instead?

I'm curious what is show by 'print/x' of the various values in there.

Luis> Also, for 32-bit targets, I'm seeing size-related issues.

We're also doing testing on ARM and that seems fine.

I did disable some of these internally for some targets, ones with
"weird" floating point.

Anyway, it seems like we could change the test to verify that the size
of each floating point type is what's expected before proceeding.  But
I'd like to understand the above a bit better, first.

thanks,
Tom

Re: [PATCH 2/3] Implement real literal extension for Ada

[Luis Machado]

Hi Tom,

On 3/18/22 18:05, Tom Tromey wrote:
>>>>>> "Luis" == Luis Machado <luis.machado@arm.com> writes:
> 
> Luis> I think the floating point representations may differ between
> Luis> architectures/bit-sizes.
> 
> Luis> I noticed 5 failures in this test for aarch64-linux and arm-linux.
> 
> This is interesting, because internally we have a rewrite of this same
> test, and it seems fine on aarch64-linux.
> 
> Luis> FAIL: gdb.ada/float-bits.exp: print 16llf#7FFFF7FF4054A56FA5B99019A5C8#
> Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double
> Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double :=
> Luis> 16llf#7FFFF7FF4054A56FA5B99019A5C8#
> Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double after assignment
> Luis> FAIL: gdb.ada/float-bits.exp: print
> 
> Luis> Some of the expected values don't really match. If the output is
> Luis> supposed to differ, maybe we should compare hex values instead?
> 
> I'm curious what is show by 'print/x' of the various values in there.

Here's what I see on my end:

(gdb) p/x 16llf#7FFFF7FF4054A56FA5B99019A5C8#
$9 = 0x7ffff7ff4054a56fa5b99019a5c8

(gdb) p/x val_long_double
$10 = 0x4544adf4b7320335

(gdb) p/x val_long_double := 16llf#7FFFF7FF4054A56FA5B99019A5C8#
$11 = 0x0

(gdb) p/x val_long_double
$12 = 0x0

(gdb) p/x 16llf#a56fa5b99019a5c800007ffff7ff4054#
$13 = 0xa56fa5b99019a5c800007ffff7ff4054

(gdb) ptype val_long_double
type = <8-byte float>

(gdb) ptype 16llf#a56fa5b99019a5c800007ffff7ff4054#
type = <16-byte float>

> 
> Luis> Also, for 32-bit targets, I'm seeing size-related issues.
> 
> We're also doing testing on ARM and that seems fine.
> 

Interesting. I wonder what's different in my environment then. This is 
standard Ubuntu 20.04 on aarch64 hardware.

I usually build --enable-targets=all, but in this case I tried a single 
target and it did not change things.

> I did disable some of these internally for some targets, ones with
> "weird" floating point.
> 
> Anyway, it seems like we could change the test to verify that the size
> of each floating point type is what's expected before proceeding.  But
> I'd like to understand the above a bit better, first.

Please let me know if you'd like the full log and/or some more information.

Re: [PATCH 2/3] Implement real literal extension for Ada

[Luis Machado]

Hi,

On 3/21/22 09:28, Luis Machado via Gdb-patches wrote:
> Hi Tom,
> 
> On 3/18/22 18:05, Tom Tromey wrote:
>>>>>>> "Luis" == Luis Machado <luis.machado@arm.com> writes:
>>
>> Luis> I think the floating point representations may differ between
>> Luis> architectures/bit-sizes.
>>
>> Luis> I noticed 5 failures in this test for aarch64-linux and arm-linux.
>>
>> This is interesting, because internally we have a rewrite of this same
>> test, and it seems fine on aarch64-linux.
>>
>> Luis> FAIL: gdb.ada/float-bits.exp: print 
>> 16llf#7FFFF7FF4054A56FA5B99019A5C8#
>> Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double
>> Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double :=
>> Luis> 16llf#7FFFF7FF4054A56FA5B99019A5C8#
>> Luis> FAIL: gdb.ada/float-bits.exp: print val_long_double after 
>> assignment
>> Luis> FAIL: gdb.ada/float-bits.exp: print
>>
>> Luis> Some of the expected values don't really match. If the output is
>> Luis> supposed to differ, maybe we should compare hex values instead?
>>
>> I'm curious what is show by 'print/x' of the various values in there.
> 
> Here's what I see on my end:
> 
> (gdb) p/x 16llf#7FFFF7FF4054A56FA5B99019A5C8#
> $9 = 0x7ffff7ff4054a56fa5b99019a5c8
> 
> (gdb) p/x val_long_double
> $10 = 0x4544adf4b7320335
> 
> (gdb) p/x val_long_double := 16llf#7FFFF7FF4054A56FA5B99019A5C8#
> $11 = 0x0
> 
> (gdb) p/x val_long_double
> $12 = 0x0
> 
> (gdb) p/x 16llf#a56fa5b99019a5c800007ffff7ff4054#
> $13 = 0xa56fa5b99019a5c800007ffff7ff4054
> 
> (gdb) ptype val_long_double
> type = <8-byte float>
> 
> (gdb) ptype 16llf#a56fa5b99019a5c800007ffff7ff4054#
> type = <16-byte float>
> 
>>
>> Luis> Also, for 32-bit targets, I'm seeing size-related issues.
>>
>> We're also doing testing on ARM and that seems fine.
>>
> 
> Interesting. I wonder what's different in my environment then. This is 
> standard Ubuntu 20.04 on aarch64 hardware.
> 
> I usually build --enable-targets=all, but in this case I tried a single 
> target and it did not change things.
> 
>> I did disable some of these internally for some targets, ones with
>> "weird" floating point.
>>
>> Anyway, it seems like we could change the test to verify that the size
>> of each floating point type is what's expected before proceeding.  But
>> I'd like to understand the above a bit better, first.
> 
> Please let me know if you'd like the full log and/or some more information.

Did you have a chance to check things on your end?