Re: [PATCH] libgccjit: add some reflection functions in the jit C api

[David Malcolm <dmalcolm@redhat.com>]

On Thu, 2020-10-15 at 13:39 -0400, Antoni Boucher wrote:
> Thanks. I updated the patch with these changes.

Thanks for patch; review below.  Sorry if it seems excessively nitpicky
in places.

> 2020-09-1  Antoni Boucher  <bouanto@zoho.com>
> 
>     gcc/jit/
>             PR target/96889
>             * docs/topics/compatibility.rst (LIBGCCJIT_ABI_14): New ABI tag.

15 now.

>             * docs/topics/functions.rst: Add documentation for the
>             functions gcc_jit_function_get_return_type and
>             gcc_jit_function_get_param_count
>             * libgccjit.c: New functions:
>               * gcc_jit_function_get_return_type;
>               * gcc_jit_function_get_param_count;
>               * gcc_jit_function_type_get_return_type;
>               * gcc_jit_function_type_get_param_count;
>               * gcc_jit_function_type_get_param_type;
>               * gcc_jit_type_unqualified;
>               * gcc_jit_type_is_array;
>               * gcc_jit_type_is_bool;
>               * gcc_jit_type_is_function_ptr_type;
>               * gcc_jit_type_is_int;
>               * gcc_jit_type_is_pointer;
>               * gcc_jit_type_is_vector;
>               * gcc_jit_vector_type_get_element_type;
>               * gcc_jit_vector_type_get_num_units;
>               * gcc_jit_struct_get_field;
>               * gcc_jit_type_is_struct;
>               * gcc_jit_struct_get_field_count;

This isn't valid ChangeLog format; it will fail the git hooks.

>             * libgccjit.h

Likewise.

>             * jit-recording.h: New functions (is_struct and is_vector)
>             * libgccjit.map (LIBGCCJIT_ABI_14): New ABI tag.

15 now.

> 
>     gcc/testsuite/
>             PR target/96889
>             * jit.dg/all-non-failing-tests.h: Add test-reflection.c.
>             * jit.dg/test-reflection.c: New test.

[...]


> diff --git a/gcc/jit/docs/topics/functions.rst b/gcc/jit/docs/topics/functions.rst
> index eb40d64010e..9819c28cda2 100644
> --- a/gcc/jit/docs/topics/functions.rst
> +++ b/gcc/jit/docs/topics/functions.rst
> @@ -171,6 +171,16 @@ Functions
>     underlying string, so it is valid to pass in a pointer to an on-stack
>     buffer.
>  
> +.. function::  size_t \
> +               gcc_jit_function_get_param_count (gcc_jit_function *func)
> +
> +   Get the number of parameters of the function.
> +
> +.. function::  gcc_jit_type \*
> +               gcc_jit_function_get_return_type (gcc_jit_function *func)
> +
> +   Get the return type of the function.
> +

The documentation part of the patch is incomplete: it hasn't been
updated to add all the new entrypoints.
Also, the return type of gcc_jit_function_get_param_count is
inconsistent (size_t above, but ssize_t below).


> diff --git a/gcc/jit/jit-recording.h b/gcc/jit/jit-recording.h
> index 30e37aff387..525b8bc921d 100644
> --- a/gcc/jit/jit-recording.h
> +++ b/gcc/jit/jit-recording.h
> @@ -538,7 +538,9 @@ public:
>    virtual bool is_bool () const = 0;
>    virtual type *is_pointer () = 0;
>    virtual type *is_array () = 0;
> +  virtual struct_ *is_struct () { return NULL; }

Can't you use dyn_cast_struct for this?
Or is this about looking through decorated_type? e.g. for const and
volatile variants?

I guess my question is, what is the purpose of gcc_jit_type_is_struct?

>    virtual bool is_void () const { return false; }
> +  virtual vector_type *is_vector () { return NULL; }

Likewise, can't you use dyn_cast_vector_type for this?

>    virtual bool has_known_size () const { return true; }
>  
>    bool is_numeric () const
> @@ -595,6 +597,8 @@ public:
>    bool is_bool () const FINAL OVERRIDE;
>    type *is_pointer () FINAL OVERRIDE { return dereference (); }
>    type *is_array () FINAL OVERRIDE { return NULL; }
> +  vector_type *is_vector () FINAL OVERRIDE { return NULL; }
> +  struct_ *is_struct () FINAL OVERRIDE { return NULL; }

Likewise, and this is redundant, as it's merely copying the base class
implementation.

>    bool is_void () const FINAL OVERRIDE { return m_kind == GCC_JIT_TYPE_VOID; }
>  
>  public:
> @@ -629,6 +633,8 @@ public:
>    bool is_bool () const FINAL OVERRIDE { return false; }
>    type *is_pointer () FINAL OVERRIDE { return m_other_type; }
>    type *is_array () FINAL OVERRIDE { return NULL; }
> +  vector_type *is_vector () FINAL OVERRIDE { return NULL; }
> +  struct_ *is_struct () FINAL OVERRIDE { return NULL; }

Likewise.


> @@ -655,6 +661,7 @@ public:
>    bool is_bool () const FINAL OVERRIDE { return m_other_type->is_bool (); }
>    type *is_pointer () FINAL OVERRIDE { return m_other_type->is_pointer (); }
>    type *is_array () FINAL OVERRIDE { return m_other_type->is_array (); }
> +  struct_ *is_struct () FINAL OVERRIDE { return m_other_type->is_struct (); }

Aha: with a decorated type you look through the decoration.

>  protected:
>    type *m_other_type;
> @@ -737,6 +744,8 @@ public:
>  
>    void replay_into (replayer *) FINAL OVERRIDE;
>  
> +  vector_type *is_vector () FINAL OVERRIDE { return this; }
> +
>  private:
>    string * make_debug_string () FINAL OVERRIDE;
>    void write_reproducer (reproducer &r) FINAL OVERRIDE;
> @@ -765,6 +774,8 @@ class array_type : public type
>    bool is_bool () const FINAL OVERRIDE { return false; }
>    type *is_pointer () FINAL OVERRIDE { return NULL; }
>    type *is_array () FINAL OVERRIDE { return m_element_type; }
> +  vector_type *is_vector () FINAL OVERRIDE { return NULL; }
> +  struct_ *is_struct () FINAL OVERRIDE { return NULL; }

Are these redundant?

>    int num_elements () { return m_num_elements; }
>  
>    void replay_into (replayer *) FINAL OVERRIDE;
> @@ -799,6 +810,8 @@ public:
>    bool is_bool () const FINAL OVERRIDE { return false; }
>    type *is_pointer () FINAL OVERRIDE { return NULL; }
>    type *is_array () FINAL OVERRIDE { return NULL; }
> +  vector_type *is_vector () FINAL OVERRIDE { return NULL; }
> +  struct_ *is_struct () FINAL OVERRIDE { return NULL; }

Likewise.

>    void replay_into (replayer *) FINAL OVERRIDE;
>  
> @@ -912,6 +925,7 @@ public:
>    bool is_bool () const FINAL OVERRIDE { return false; }
>    type *is_pointer () FINAL OVERRIDE { return NULL; }
>    type *is_array () FINAL OVERRIDE { return NULL; }
> +  vector_type *is_vector () FINAL OVERRIDE { return NULL; }

Likewise.

>    bool has_known_size () const FINAL OVERRIDE { return m_fields != NULL; }
>  
> @@ -943,6 +957,8 @@ public:
>  
>    const char *access_as_type (reproducer &r) FINAL OVERRIDE;
>  
> +  struct_ *is_struct () FINAL OVERRIDE { return this; }
> +
>  private:
>    string * make_debug_string () FINAL OVERRIDE;
>    void write_reproducer (reproducer &r) FINAL OVERRIDE;
> diff --git a/gcc/jit/libgccjit.c b/gcc/jit/libgccjit.c
> index a00aefc7108..8e6a9e85ff0 100644
> --- a/gcc/jit/libgccjit.c
> +++ b/gcc/jit/libgccjit.c
> @@ -21,6 +21,7 @@ along with GCC; see the file COPYING3.  If not see
>  #include "config.h"
>  #include "system.h"
>  #include "coretypes.h"
> +#include "tm.h"

Why is this included?  It's not mentioned in the ChangeLog.

>  #include "timevar.h"
>  #include "typed-splay-tree.h"
>  #include "cppbuiltin.h"
> @@ -60,6 +61,14 @@ struct gcc_jit_struct : public gcc::jit::recording::struct_
>  {
>  };
>  
> +struct gcc_jit_function_type : public gcc::jit::recording::function_type
> +{
> +};
> +
> +struct gcc_jit_vector_type : public gcc::jit::recording::vector_type
> +{
> +};
> +

FWIW these aren't mentioned in the ChangeLog either.

>  struct gcc_jit_field : public gcc::jit::recording::field
>  {
>  };
> @@ -510,6 +519,197 @@ gcc_jit_type_get_volatile (gcc_jit_type *type)

[...]

> +/* Public entrypoint.  See description in libgccjit.h.
> +
> +   After error-checking, the real work is done by the
> +   gcc::jit::recording::function_type::get_param_types method, in
> +   jit-recording.c.  */
> +
> +gcc_jit_type *
> +gcc_jit_function_type_get_param_type (gcc_jit_function_type *function_type,
> +				int index)
> +{
> +  RETURN_NULL_IF_FAIL (function_type, NULL, NULL, "NULL struct_type");
> +  int num_params = function_type->get_param_types ().length ();
> +  gcc::jit::recording::context *ctxt = function_type->m_ctxt;

Should check that index >= 0 as well (or make it unsigned).

Given that gcc_jit_function_get_param_count returns ssize_t, should
that be the type of "index"?

> +  RETURN_NULL_IF_FAIL_PRINTF3 (index < num_params,
> +			       ctxt, NULL,
> +			       "index of %d is too large (%s has %d params)",
> +			       index,
> +			       function_type->get_debug_string (),
> +			       num_params);
> +  return (gcc_jit_type *)function_type->get_param_types ()[index];
> +}

[...]

> +extern gcc_jit_field *
> +gcc_jit_struct_get_field (gcc_jit_struct *struct_type,
> +			   int index)
> +{
> +  RETURN_NULL_IF_FAIL (struct_type, NULL, NULL, "NULL struct type");

Similar comments as per gcc_jit_function_type_get_param_type above.
Should check here that index is in range (>=0 and < num fields).

> +  return (gcc_jit_field *)struct_type->get_fields ()->get_field (index);
> +}
> +
> +/* Public entrypoint.  See description in libgccjit.h.
> +
> +   After error-checking, this calls the trivial
> +   gcc::jit::recording::struct_::get_fields method in
> +   jit-recording.h.  */
> +
> +ssize_t
> +gcc_jit_struct_get_field_count (gcc_jit_struct *struct_type)
> +{
> +  RETURN_VAL_IF_FAIL (struct_type, -1, NULL, NULL, "NULL struct type");
> +  return struct_type->get_fields ()->length ();
> +}
> +

[...]

> diff --git a/gcc/jit/libgccjit.h b/gcc/jit/libgccjit.h
> index 7134841bb07..6056d00a3ea 100644
> --- a/gcc/jit/libgccjit.h
> +++ b/gcc/jit/libgccjit.h
> @@ -96,6 +96,12 @@ typedef struct gcc_jit_field gcc_jit_field;
>     the layout for, or an opaque type.  */
>  typedef struct gcc_jit_struct gcc_jit_struct;
>  
> +/* A gcc_jit_function_type encapsulates a function type.  */
> +typedef struct gcc_jit_function_type gcc_jit_function_type;
> +
> +/* A gcc_jit_vector_type encapsulates a vector type.  */
> +typedef struct gcc_jit_vector_type gcc_jit_vector_type;
> +
>  /* A gcc_jit_function encapsulates a function: either one that you're
>     creating yourself, or a reference to one that you're dynamically
>     linking to within the rest of the process.  */
> @@ -586,6 +592,61 @@ gcc_jit_type_get_const (gcc_jit_type *type);
>  extern gcc_jit_type *
>  gcc_jit_type_get_volatile (gcc_jit_type *type);
>  
> +/* Given type "T", return TRUE if the type is an array.  */
> +extern gcc_jit_type *
> +gcc_jit_type_is_array (gcc_jit_type *type);

The comment is unclear; does this actually return the type vs NULL?

As noted above, I think we need to think about what happens on
qualified types; is stripping them the responsibility of the caller?
Should it be?  (I'm thinking "yes", but presumably all this is
motivated by your rust hacking).

Also, there isn't a type "T" here; the param name is "type", and there
isn't a supporting example using "T" as a placeholder.
Maybe just lose the 'Given type "T", ' prefix here and below (unless
giving a example).

> +/* Given type "T", return TRUE if the type is a bool.  */
> +extern int
> +gcc_jit_type_is_bool (gcc_jit_type *type);

I'd prefer "return non-zero" over "return TRUE" (here, and in various
places below).

Does this check for the "bool" type i.e.:
  /* C++'s bool type; also C99's "_Bool" type, aka "bool" if using
     stdbool.h.  */
  GCC_JIT_TYPE_BOOL,
?

> +/* Given type "T", return the function type if it is one.  */

..."or NULL".

> +extern gcc_jit_function_type *
> +gcc_jit_type_is_function_ptr_type (gcc_jit_type *type);
> +
> +/* Given a function type, return its return type.  */
> +extern gcc_jit_type *
> +gcc_jit_function_type_get_return_type (gcc_jit_function_type *function_type);
> +
> +/* Given a function type, return its number of parameters.  */
> +extern ssize_t
> +gcc_jit_function_type_get_param_count (gcc_jit_function_type *function_type);
> +
> +/* Given a function type, return the type of the specified parameter.  */
> +extern gcc_jit_type *
> +gcc_jit_function_type_get_param_type (gcc_jit_function_type *function_type,
> +				int index);
> +
> +/* Given type "T", return the type pointed by the pointer type
> + * or NULL if it's not a pointer.  */
> +extern gcc_jit_type *
> +gcc_jit_type_is_pointer (gcc_jit_type *type);
> +
> +/* Given type "T", return TRUE if the type is an int.  */
> +extern int
> +gcc_jit_type_is_int (gcc_jit_type *type);

I'm not sure I understand what this function does, which suggests it's
misnamed.  The current name suggests that it checks for the C "int"
type, whereas I believe it actually checks for an integral
type.  Should it be:

extern int
gcc_jit_type_is_integral (gcc_jit_type *type);

?

What's the output?  Do you merely need nonzero/zero for boolean, or
information on the width of the type/signedness, or whatnot?  (perhaps
have some out params for extracting width/signedness for when it
returns nonzero?)

Looking at the testcase, it looks like it currently returns false on
the _Bool type.  Is that correct?


> +/* Given type "T", return the unqualified type, i.e. for a
> + * decorated type, return the type it wraps.  */
> +extern gcc_jit_type *
> +gcc_jit_type_unqualified (gcc_jit_type *type);

"decorated" is a term used inside the implementation, but not in the
user-facing docs.

How about:

  /* Given a type, return the unqualified type, removing "const",
"volatile"
     and alignment qualifiers.  */

or somesuch.

> +/* Given type "T", return TRUE if the type is a vector type.  */
> +extern gcc_jit_vector_type *
> +gcc_jit_type_is_vector (gcc_jit_type *type);

As above, this doesn't return TRUE, it's more of a dynamic cast.
Likewise about responsibility for qualifiers.

> +/* Given type "T", return the struct type or NULL.  */
> +extern gcc_jit_struct *
> +gcc_jit_type_is_struct (gcc_jit_type *type);

Wording could be improved as noted above.

> +/* Given a vector type, return the number of units it contains.  */
> +extern ssize_t
> +gcc_jit_vector_type_get_num_units (gcc_jit_vector_type *vector_type);
> +
> +/* Given a vector type, return the type of its elements.  */
> +extern gcc_jit_type *
> +gcc_jit_vector_type_get_element_type (gcc_jit_vector_type *vector_type);
> +

[...]

> @@ -647,6 +708,15 @@ gcc_jit_struct_set_fields (gcc_jit_struct *struct_type,
>  			   int num_fields,
>  			   gcc_jit_field **fields);
>  
> +/* Get a field by index.  */
> +extern gcc_jit_field *
> +gcc_jit_struct_get_field (gcc_jit_struct *struct_type,
> +			   int index);
> +
> +/* Get the number of fields.  */
> +extern ssize_t
> +gcc_jit_struct_get_field_count (gcc_jit_struct *struct_type);

Do we need these for unions?  Or skip them for now?

> @@ -740,6 +810,14 @@ gcc_jit_function_as_object (gcc_jit_function *func);
>  extern gcc_jit_param *
>  gcc_jit_function_get_param (gcc_jit_function *func, int index);
>  
> +/* Get the number of params of a function.  */
> +extern ssize_t
> +gcc_jit_function_get_param_count (gcc_jit_function *func);
> +
> +/* Get the return type of a function.  */
> +extern gcc_jit_type *
> +gcc_jit_function_get_return_type (gcc_jit_function *func);

Are these redundant if there's a way to ask a gcc_jit_function for its
type, and then to use the gcc_jit_function_type_ functions?
(I don't mind if they are)

> @@ -1518,6 +1596,47 @@ gcc_jit_version_minor (void);
>  extern int
>  gcc_jit_version_patchlevel (void);
>  
> +#define LIBGCCJIT_HAVE_gcc_jit_function_reflection
> +
> +/* Reflection functions to get the number of parameters, return type of
> +   a function and whether a type is a bool from the C API.
> +
> +   This API entrypoint was added in LIBGCCJIT_ABI_15; you can test for its
> +   presence using
> +     #ifdef LIBGCCJIT_HAVE_gcc_jit_function_reflection
> +*/

This has become a big mixture of things beyond just function
reflection, so I think I'd prefer it if each entrypoint "funcname" had
its own "LIBGCCJIT_HAVE_funcname" define.  Or just "REFLECTION" rather
than "function_reflection", that's probably simpler.
Note that the existing "HAVE_foo" defines are lowercase "foo" when it's
an actual entrypoint named "foo", and are uppercase otherwise e.g.
LIBGCCJIT_HAVE_SWITCH_STATEMENTS and LIBGCCJIT_HAVE_TIMING_API.

> +extern gcc_jit_type *
> +gcc_jit_function_get_return_type (gcc_jit_function *func);
> +extern ssize_t
> +gcc_jit_function_get_param_count (gcc_jit_function *func);
> +extern gcc_jit_type *
> +gcc_jit_type_is_array (gcc_jit_type *type);
> +extern int
> +gcc_jit_type_is_bool (gcc_jit_type *type);
> +extern gcc_jit_function_type *
> +gcc_jit_type_is_function_ptr_type (gcc_jit_type *type);
> +extern gcc_jit_type *
> +gcc_jit_function_type_get_return_type (gcc_jit_function_type *function_type);
> +extern ssize_t
> +gcc_jit_function_type_get_param_count (gcc_jit_function_type *function_type);
> +extern gcc_jit_type *
> +gcc_jit_function_type_get_param_type (gcc_jit_function_type *function_type,
> +				int index);
> +extern int
> +gcc_jit_type_is_int (gcc_jit_type *type);
> +extern gcc_jit_type *
> +gcc_jit_type_is_pointer (gcc_jit_type *type);
> +extern gcc_jit_vector_type *
> +gcc_jit_type_is_vector (gcc_jit_type *type);
> +extern gcc_jit_struct *
> +gcc_jit_type_is_struct (gcc_jit_type *type);
> +extern ssize_t
> +gcc_jit_vector_type_get_num_units (gcc_jit_vector_type *vector_type);
> +extern gcc_jit_type *
> +gcc_jit_vector_type_get_element_type (gcc_jit_vector_type *vector_type);
> +extern gcc_jit_type *
> +gcc_jit_type_unqualified (gcc_jit_type *type);

Is this part of the patch a duplicate copy of the various things added
above?  Perhaps copy-and-paste error, or am I misreading this?

[...]

Hope this is constructive
Dave