[tarantool-patches] Re: [PATCH v1 1/1] lua: export key_def methods for LUA key_def object

[Alexander Turenko]

I would prefer if you will answer to review comments explicitly.

Re using lua_tolstring(): I understood that lua_tolstring() only
guarantees that a string is valid until it will be popped from a stack
(despite that fact that we know it has another copy on the stack) and so
a region usage looks good.

Just for the record: We discussed voicely that make this patch depending
on the luaT_tuple_new() implementation from the merger patchset is not
worth to do (just saves two lines of code here).

Changes I made (I mostly add them into a separate 'FIXUP' commit to
allow you to look into them if you want; pushed to your branch back):

* Squashed the commits, updated the commit message.
* Renamed :to_table() to totable().
* Use memcpy(tmp, path, path_len + 1) instead of an explicit '\0'
  assignment (lua_tolstring() allows it).
* Use 'struct lua_State' instead of just 'lua_State' to better match
  with our code style.
* Rewrote a test: split it by cases, do some renaming, fixed mine
  parts = {...} to parts = {{...}}, added more cases.
* Found an assertion fault with a new test case:

tarantool: /home/alex/p/tarantool-meta/r/tarantool/src/box/tuple_extract_key.cc:140: char* tuple_extract_key_slowpath(const tuple*, key_def*, uint32_t*) [with bool contains_sequential_parts = false; bool has_optional_parts = false; bool has_json_paths = false; uint32_t = unsigned int]: Assertion `field != NULL' failed.
Aborted

See the first commented test case. I think that we should allow a
fieldno in key_def that is above then a tuple length when the field is
nullable and raise an error otherwise (when it is not nullable). (I
guess that we can do it at least partially by setting
has_optional_parts).

However when I tried with non-nullable part with {fieldno = 2}
:extract_key({'foo'}) gives me {'foo', 80} -- it seems to go over its
memory. But it asserts like the first one if I added {fieldno = 3}.

All that cases are commented now. Can you please elaborate how to
archieve the behaviour I proposed above and whether it is good way to
go, Kirill?

Also I have an idea: maybe add ability to use tables as tuples? We have
to receive tuples in this way from net.box.

WBR, Alexander Turenko.

On Tue, Mar 26, 2019 at 04:37:54PM +0300, Kirill Shcherbatov wrote:
> Hi! Thank you for review.
> https://github.com/tarantool/tarantool/tree/kshch/gh-4025-lua-key-kef-methods
> https://github.com/tarantool/tarantool/tree/kshch/gh-4025-lua-key-kef-methods-110