Exposing more types to native modules #5641

Activity

1. dpgeorge commented on Feb 14, 2020

   

   dpgeorge

   on Feb 14, 2020

   Member

   More actions

   You need to make sure the changes are reflected in this table: https://github.com/micropython/micropython/blob/master/tools/mpy_ld.py#L666-L682
2. dpgeorge

   added

   py-coreRelates to py/ directory in sourceRelates to py/ directory in source

   on Feb 14, 2020
3. jeremyherbert commented on Feb 14, 2020

   

   jeremyherbert

   on Feb 14, 2020

   ContributorAuthor

   More actions

   Yes, I've added the symbols to that (unfortunately it doesn't fix the issue, so I think I have a problem elsewhere).

   I would have thought that without them I should hit the LinkError on line 704 anyway?
4. jeremyherbert commented on Feb 14, 2020

   

   jeremyherbert

   on Feb 14, 2020

   Last edited by jeremyherbert

   ContributorAuthor

   More actions

   Just some more notes for me to refer to later:

   after building micropython with DEBUG=1, pahole shows that the exported symbol table has the correct structure:

   pahole --reorganize -C _mp_fun_table_t micropython gives

   ...
   	mp_obj_t                   (*obj_new_str)(const char  *, size_t); /*   456     8 */
   	mp_obj_t                   (*obj_new_bytes)(const byte  *, size_t); /*   464     8 */
   	mp_obj_t                   (*obj_new_bytearray_by_ref)(size_t, void *); /*   472     8 */
   	mp_obj_t                   (*obj_new_float_from_f)(float); /*   480     8 */
   	mp_obj_t                   (*obj_new_float_from_d)(double); /*   488     8 */
   	float                      (*obj_get_float_to_f)(mp_obj_t); /*   496     8 */
   	double                     (*obj_get_float_to_d)(mp_obj_t); /*   504     8 */
   	/* --- cacheline 8 boundary (512 bytes) --- */
   	void                       (*get_buffer_raise)(mp_obj_t, mp_buffer_info_t *, mp_uint_t); /*   512     8 */
   	const mp_stream_p_t  *     (*get_stream_raise)(mp_obj_t, int); /*   520     8 */
   	const mp_print_t  *        plat_print;           /*   528     8 */
   	const mp_obj_type_t  *     type_type;            /*   536     8 */
   	const mp_obj_type_t  *     type_int;             /*   544     8 */
   	const mp_obj_type_t  *     type_float;           /*   552     8 */
   	const mp_obj_type_t  *     type_str;             /*   560     8 */
   	const mp_obj_type_t  *     type_bytes;           /*   568     8 */
   	/* --- cacheline 9 boundary (576 bytes) --- */
   	const mp_obj_type_t  *     type_list;            /*   576     8 */
   	const mp_obj_type_t  *     type_dict;            /*   584     8 */
   	const mp_obj_type_t  *     type_fun_builtin_0;   /*   592     8 */
   	const mp_obj_type_t  *     type_fun_builtin_1;   /*   600     8 */
   	const mp_obj_type_t  *     type_fun_builtin_2;   /*   608     8 */
   	const mp_obj_type_t  *     type_fun_builtin_3;   /*   616     8 */
   	const mp_obj_type_t  *     type_fun_builtin_var; /*   624     8 */
   	const mp_obj_fun_builtin_var_t  * stream_read_obj; /*   632     8 */
   ...
   

   And indeed the table is still lined up with the mp_ld.py offset of 67 * 8 = mp_type_type.
5. jeremyherbert commented on Feb 14, 2020

   

   jeremyherbert

   on Feb 14, 2020

   Last edited by jeremyherbert

   ContributorAuthor

   More actions

   modifying mp_ld.py to print out symbol offsets shows that mp_type_fun_builtin_1 is indeed being mapped to offset 576 in the native module, rather than the correct offset of 600.

   But in this process I actually just found my dumb mistake, see if you can spot it:

       mp_fun_table_sec = Section('.external.mp_fun_table', b'', 0)
       fun_table = {key: 67 + idx
           for idx, key in enumerate([
               'mp_type_type',
               'mp_type_int',
               'mp_type_float'
               'mp_type_str',
               'mp_type_bytes'
               'mp_type_list',
               'mp_type_dict',
               'mp_type_fun_builtin_0',
               'mp_type_fun_builtin_1',
               'mp_type_fun_builtin_2',
               'mp_type_fun_builtin_3',
               'mp_type_fun_builtin_var',
               'mp_stream_read_obj',
               'mp_stream_readinto_obj',
               'mp_stream_unbuffered_readline_obj',
               'mp_stream_write_obj',
           ])
       }
   
6. jeremyherbert

   closed this as completedon Feb 14, 2020
7. dpgeorge commented on Feb 14, 2020

   

   dpgeorge

   on Feb 14, 2020

   Member

   More actions

   But in this process I actually just found my dumb mistake, see if you can spot it:

   Ok, so you missed commas!

   It might be a good idea to generate this table in mpy_ld.py automatically (when that script runs) by parsing py/nativeglue.h.
8. jeremyherbert commented on Feb 14, 2020

   

   jeremyherbert

   on Feb 14, 2020

   ContributorAuthor

   More actions

   Ah, didn't see this in time. I just created a PR with all of the types from obj.h.

   I'll need to think a bit about parsing, because from memory pycparsing doesn't like macros and micropython obviously uses them a lot.
9. tannewt

   added a commit that references this issue on Dec 2, 2021

   Merge pull request micropython#5641 from jepler/bitbangio-spi-write-s…

   ...

   39cc38e