lua: prevent serialization of error for ucdata

[Buristan]

The alternative solution is to use just lua_call() instead, since the error should be caught somewhere in the upper frames.

diff --git a/src/lua/serializer.c b/src/lua/serializer.c
index 1018f9572..56400e7f3 100644
--- a/src/lua/serializer.c
+++ b/src/lua/serializer.c
@@ -274,7 +274,11 @@ lua_field_inspect_ucdata(struct lua_State *L, struct luaL_serializer *cfg,
                 }
                 /* copy object itself */
                 lua_pushvalue(L, idx);
-                lua_pcall(L, 1, 1, 0);
+                /*
+                 * The possible error should be catched in some
+                 * upper frame.
+                 */
+                lua_call(L, 1, 1);
                 /* replace obj with the unpacked value */
                 lua_replace(L, idx);
                 if (luaL_tofield(L, cfg, idx, field) < 0)

The results of the benchmark below for both ways are quite similar, so I prefer the most common way (the one in the patch):

time src/tarantool -e "                                                                
local yml = require'yaml'
local ud_mt = {__serialize = function()
  return ''
end}
ud_mt.__index = ud_mt
local ud = newproxy(true)
debug.setmetatable(ud, ud_mt)
collectgarbage('stop')
for _ = 1, 3e6 do
    yml.encode(ud)
end
"  

lua_pcall()	lua_call()
$4.896s\ (\sigma=0.037$)	$4.879s\ (\sigma=0.034$)

[coveralls]

coverage: 86.688% (+0.04%) from 86.644%
when pulling 2a57129 on Buristan:skaplun/gh-9396-ucdata-serialize-error
into 3ee68d8
on tarantool:master.

[igormunkin]

@Buristan, thanks for the patch! LGTM.

[Lord-KA]

Thanks for the patch. I don't see any issues with the change, LGTM.

Nit: @Buristan, I am just curious, why did you leave the XXX in the test comments?

[Buristan]

@Lord-KA,
AFAIK, XXX is just a legacy keyword coming from Sun code convention:

10.5.4 Special Comments
Use XXX in a comment to flag something that is bogus but works. Use FIXME to flag something that is bogus and broken.

[Lord-KA]

Personally, I don't see anything wrong with using yaml.encode() in this reproducer, but sure.

[igormunkin]

Also backported to 2.11 in scope of 074fe0b. To apply the changes for 2.10, @Buristan will create a separate PR with the corresponding updates.