[PATCH] [gdbsupport] Poison ctype.h / cctype functions

[PATCH] [gdbsupport] Poison ctype.h / cctype functions

[Tom de Vries]

GDB generally uses c_isprint instead of isprint or std::isprint [1].

Poison uses of ctype.h / cctype functions in gdbsupport/poison.h.

This excludes isupper, islower and toupper because they're currently used in
GDB.  Add a fixme to find out whether these exceptions are ok, and if so
document why, otherwise fix them.

Tested on x86_64-linux.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=34017

[1] https://sourceware.org/pipermail/gdb-patches/2026-March/226117.html
---
 gdbsupport/poison.h | 52 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 52 insertions(+)

diff --git a/gdbsupport/poison.h b/gdbsupport/poison.h
index 904646110a5..0963645ee81 100644
--- a/gdbsupport/poison.h
+++ b/gdbsupport/poison.h
@@ -237,4 +237,56 @@ non-POD data type.");
 #undef XOBNEWVEC
 #define XOBNEWVEC(O, T, N) xobnewvec<T> (O, N)
 
+#ifndef IN_PROCESS_AGENT
+/* Poison functions like isprint from header filer ctype.h and std::isprint
+   from header file cctype.  */
+
+#define POISON_CCTYPE(F)					\
+  template<typename T>						\
+  F								\
+  {								\
+    static_assert (false, "Please use gnulib's c-ctype.h");	\
+    return 0;							\
+  }
+
+/* Poisoning of some functions is disabled:
+   - isupper, because it's used in ada_decode.
+   - islower and toupper, because they're used in UNCTRL from readline,
+     which is used in tui_redisplay_readline and gdb_fnprint.
+
+   FIXME: Find out whether these exceptions are ok, and if so document
+   why, otherwise fix them.  */
+
+POISON_CCTYPE (int isalnum (T c));
+POISON_CCTYPE (int std::isalnum (T c));
+POISON_CCTYPE (int isalpha (T c));
+POISON_CCTYPE (int std::isalpha (T c));
+POISON_CCTYPE (int isblank (T c));
+POISON_CCTYPE (int std::isblank (T c));
+POISON_CCTYPE (int iscntrl (T c));
+POISON_CCTYPE (int std::iscntrl (T c));
+POISON_CCTYPE (int isdigit (T c));
+POISON_CCTYPE (int std::isdigit (T c));
+POISON_CCTYPE (int isgraph (T c));
+POISON_CCTYPE (int std::isgraph (T c));
+/* Disabled: int islower (T c)).  */
+POISON_CCTYPE (int std::islower (T c));
+POISON_CCTYPE (int isprint (T c));
+POISON_CCTYPE (int std::isprint (T c));
+POISON_CCTYPE (int ispunct (T c));
+POISON_CCTYPE (int std::ispunct (T c));
+POISON_CCTYPE (int isspace (T c));
+POISON_CCTYPE (int std::isspace (T c));
+/* Disabled: int isupper (T c).  */
+POISON_CCTYPE (int std::isupper (T c));
+POISON_CCTYPE (int isxdigit (T c));
+POISON_CCTYPE (int std::isxdigit (T c));
+POISON_CCTYPE (int tolower (T c));
+POISON_CCTYPE (int std::tolower (T c));
+/* Disabled: int toupper (T c).  */
+POISON_CCTYPE (int std::toupper (T c));
+
+#undef POISON_CCTYPE
+#endif
+
 #endif /* GDBSUPPORT_POISON_H */

base-commit: efc26507d9998eef6dc0b17480e0d921f40958c1
-- 
2.51.0

Re: [PATCH] [gdbsupport] Poison ctype.h / cctype functions

[Tom Tromey]

>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:

Tom> GDB generally uses c_isprint instead of isprint or std::isprint [1].
Tom> Poison uses of ctype.h / cctype functions in gdbsupport/poison.h.

Tom> This excludes isupper, islower and toupper because they're currently used in
Tom> GDB.  Add a fixme to find out whether these exceptions are ok, and if so
Tom> document why, otherwise fix them.

Tom> +#define POISON_CCTYPE(F)					\
Tom> +  template<typename T>						\
Tom> +  F								\
Tom> +  {								\
Tom> +    static_assert (false, "Please use gnulib's c-ctype.h");	\
Tom> +    return 0;							\
Tom> +  }
Tom> +
Tom> +/* Poisoning of some functions is disabled:
Tom> +   - isupper, because it's used in ada_decode.

This is just an oversight.

Tom> +   - islower and toupper, because they're used in UNCTRL from readline,
Tom> +     which is used in tui_redisplay_readline and gdb_fnprint.

Can't really touch readline but we could write our own UNCTRL variant.

Tom> +POISON_CCTYPE (int isalnum (T c));
Tom> +POISON_CCTYPE (int std::isalnum (T c));

I suspect redefining standard functions like this is not allowed in C++.

Tom

Re: [PATCH] [gdbsupport] Poison ctype.h / cctype functions

[Tom de Vries]

On 3/23/26 3:17 PM, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:
> 
> Tom> GDB generally uses c_isprint instead of isprint or std::isprint [1].
> Tom> Poison uses of ctype.h / cctype functions in gdbsupport/poison.h.
> 
> Tom> This excludes isupper, islower and toupper because they're currently used in
> Tom> GDB.  Add a fixme to find out whether these exceptions are ok, and if so
> Tom> document why, otherwise fix them.
> 
> Tom> +#define POISON_CCTYPE(F)					\
> Tom> +  template<typename T>						\
> Tom> +  F								\
> Tom> +  {								\
> Tom> +    static_assert (false, "Please use gnulib's c-ctype.h");	\
> Tom> +    return 0;							\
> Tom> +  }
> Tom> +
> Tom> +/* Poisoning of some functions is disabled:
> Tom> +   - isupper, because it's used in ada_decode.
> 
> This is just an oversight.
> 

I've submitted a patch for this ( 
https://sourceware.org/pipermail/gdb-patches/2026-March/226148.html ).

Thanks,
- Tom

> Tom> +   - islower and toupper, because they're used in UNCTRL from readline,
> Tom> +     which is used in tui_redisplay_readline and gdb_fnprint.
> 
> Can't really touch readline but we could write our own UNCTRL variant.
> 
> Tom> +POISON_CCTYPE (int isalnum (T c));
> Tom> +POISON_CCTYPE (int std::isalnum (T c));
> 
> I suspect redefining standard functions like this is not allowed in C++.
> 
> Tom

Re: [PATCH] [gdbsupport] Poison ctype.h / cctype functions

[Tom de Vries]

On 3/23/26 3:17 PM, Tom Tromey wrote:
> Tom> +   - islower and toupper, because they're used in UNCTRL from readline,
> Tom> +     which is used in tui_redisplay_readline and gdb_fnprint.
> 
> Can't really touch readline but we could write our own UNCTRL variant.

I've given that a try ( 
https://sourceware.org/pipermail/gdb-patches/2026-March/226171.html ).

Thanks,
- Tom

Re: [PATCH] [gdbsupport] Poison ctype.h / cctype functions

[Tom de Vries]

On 3/23/26 3:17 PM, Tom Tromey wrote:
> Tom> +POISON_CCTYPE (int isalnum (T c));
> Tom> +POISON_CCTYPE (int std::isalnum (T c));
> 
> I suspect redefining standard functions like this is not allowed in C++.

That's a pity.

Then automatically checking this falls in the domain of precommit 
checkers.  Maybe I could extend check-gnu-style to include this (though 
this isn't really a GNU style thing I suppose).

Thanks,
- Tom

Re: [PATCH] [gdbsupport] Poison ctype.h / cctype functions

[Tom Tromey]

>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:

Tom> On 3/23/26 3:17 PM, Tom Tromey wrote:
Tom> +POISON_CCTYPE (int isalnum (T c));
Tom> +POISON_CCTYPE (int std::isalnum (T c));
>> I suspect redefining standard functions like this is not allowed in
>> C++.

Tom> That's a pity.

Tom> Then automatically checking this falls in the domain of precommit
Tom> checkers.  Maybe I could extend check-gnu-style to include this
Tom> (though this isn't really a GNU style thing I suppose).

Yeah.  I considered adding a checker to avoid includes of <ctype.h>
but that doesn't actually prevent the use of these things, because they
can be made visible via other includes anyway.

Tom