Tarantool development patches archive
 help / color / mirror / Atom feed
From: Alexander Turenko <alexander.turenko@tarantool.org>
To: Igor Munkin <imun@tarantool.org>
Cc: tarantool-patches@dev.tarantool.org,
	Vladislav Shpilevoy <v.shpilevoy@tarantool.org>
Subject: Re: [Tarantool-patches] [PATCH v2 2/3] merger: clean fiber-local Lua stack after next()
Date: Fri, 17 Jul 2020 06:08:51 +0300	[thread overview]
Message-ID: <20200717030851.r2zlxwgbd4nnzcjq@tkn_work_nb> (raw)
In-Reply-To: <20200716220720.GB18920@tarantool.org>

> > > > @@ -206,14 +216,10 @@ luaT_temp_luastate(int *coro_ref)
> > > >   * It is the other half of `luaT_temp_luastate()`.
> > > >   */
> > > >  static void
> > > > -luaT_release_temp_luastate(int coro_ref)
> > > > +luaT_release_temp_luastate(struct lua_State *L, int coro_ref, int top)
> > > >  {
> > > > -	/*
> > > > -	 * FIXME: The reusable fiber-local Lua state is not
> > > > -	 * unreferenced here (coro_ref == LUA_REFNIL), but
> > > > -	 * it must be truncated to its past top to prevent
> > > > -	 * stack overflow.
> > > > -	 */
> > > > +	if (top >= 0)
> > > > +		lua_settop(L, top);
> > > >  	luaL_unref(tarantool_L, LUA_REGISTRYINDEX, coro_ref);
> > > 
> > > Minor: You can just either restore top value for fiber-local Lua state
> > > or unreference Lua coroutine without restoring a pointer to its stack
> > > top slot. As a result you need to preserve the top value only for the
> > > first case (i.e. when the coro_ref is LUA_NOREF) and ignore the value
> > > for all other cases.
> > 
> > Are you propose the following?
> > 
> >  | if (top >= 0)
> >  |         lua_settop(L, top);
> >  | else
> >  |         luaL_unref(tarantool_L, LUA_REGISTRYINDEX, coro_ref);
> 
> Kinda, but with <coro_ref> in condition:
> | if (coro_ref == LUA_NOREF)
> | 	lua_settop(L, top);
> | else
> | 	luaL_unref(tarantool_L, LUA_REGISTRYINDEX, coro_ref);
> 
> Feel free to ignore this comment if it doesn't looks logical to you.

To be honest, I would prefer to keep the current way:
luaT_temp_luastate() fills <top> and <coro_ref> and they separately
control whether lua_settop() and luaL_unref() will be called / will have
an effect in luaT_release_temp_luastate(). It is quite straightforward.

But I understand that when lua_settop() has the check, but luaL_unref()
has no, it looks a bit lopsided and it is quite natural to want to make
it visually balanced.

> > Vlad even proposed to drop `top >= 0`, which works in fact, but Lua
> > Reference Manual does not guarantee it (it was discussed within this
> > mailing thread).
> 
> I see Reference Manual allows any integer number to be passed as <idx>
> but I agree that behaviour for negative values are not clear from it.
> JFYI, LuaJIT works the same way here vanilla Lua 5.1 does. So I'm much
> closer to Vlad's opinion here, but doesn't insist on such <lua_settop>
> usage, since it might not be clear from Manual.

I'm quite sure we just have no such guarantee.

I disagree with the point that it is guaranteed, but in some unclear
way. No. I performed an attempt to find what exactly guarantees this
behaviour and found nothing. I described it in [1].

Since it unlikely will be changed, we can remove the check and leave a
comment to describe the behaviour we lean on. But the check just
shorter and cleaner, I don't see a reason to replace it with the
comment.

[1]: https://lists.tarantool.org/pipermail/tarantool-patches/2020-June/017750.html

  reply	other threads:[~2020-07-17  3:09 UTC|newest]

Thread overview: 24+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-06-17 21:06 [Tarantool-patches] [PATCH v2 0/3] Merger's NULL defererence Alexander Turenko
2020-06-17 21:06 ` [Tarantool-patches] [PATCH v2 1/3] merger: fix NULL dereference when called via iproto Alexander Turenko
2020-06-18 22:48   ` Vladislav Shpilevoy
2020-06-19  8:50     ` Alexander Turenko
2020-06-19 23:32   ` Vladislav Shpilevoy
2020-06-21 18:28     ` Alexander Turenko
2020-07-01 20:36   ` Igor Munkin
2020-07-16 20:10     ` Alexander Turenko
2020-07-16 21:42       ` Igor Munkin
2020-07-16 22:44         ` Igor Munkin
2020-07-17  3:08         ` Alexander Turenko
2020-06-17 21:06 ` [Tarantool-patches] [PATCH v2 2/3] merger: clean fiber-local Lua stack after next() Alexander Turenko
2020-06-19  8:50   ` Alexander Turenko
2020-07-01 20:36   ` Igor Munkin
2020-07-16 20:11     ` Alexander Turenko
2020-07-16 22:07       ` Igor Munkin
2020-07-17  3:08         ` Alexander Turenko [this message]
2020-06-17 21:06 ` [Tarantool-patches] [PATCH v2 3/3] lua: expose temporary Lua state for iproto calls Alexander Turenko
2020-07-01 20:37   ` Igor Munkin
2020-07-16 20:11     ` Alexander Turenko
2020-07-16 22:33       ` Igor Munkin
2020-07-17  3:09         ` Alexander Turenko
2020-06-22 20:38 ` [Tarantool-patches] [PATCH v2 0/3] Merger's NULL defererence Vladislav Shpilevoy
2020-07-17 11:28 ` Alexander Turenko

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200717030851.r2zlxwgbd4nnzcjq@tkn_work_nb \
    --to=alexander.turenko@tarantool.org \
    --cc=imun@tarantool.org \
    --cc=tarantool-patches@dev.tarantool.org \
    --cc=v.shpilevoy@tarantool.org \
    --subject='Re: [Tarantool-patches] [PATCH v2 2/3] merger: clean fiber-local Lua stack after next()' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox