fix(core): Auto cast doubles to ints in lua scripts #922
Conversation
dranikpg
force-pushed
the
fix-interpreter-double
branch
from
24e0f84 to
725c9ba
Compare
src/core/interpreter.cc
Outdated
| // Here is the use-case: | ||
| // local foo = redis.call('zscore', 'myzset', 'one') | ||
| // assert(type(foo) == "number") | ||
| // We return doubles as strings likewise to Redis to maintain compatibility. |
There was a problem hiding this comment.
can we at least use a run-time flag for this?
src/server/multi_test.cc
Outdated
| @@ -381,6 +381,10 @@ TEST_F(MultiTest, Eval) { | |||
| ASSERT_THAT(resp, ArrLen(3)); | |||
| const auto& arr = resp.GetVec(); | |||
| EXPECT_THAT(arr, ElementsAre("a", "b", "c")); | |||
|
|
|||
| Run({"zadd", "z1", "123", "a"}); | |||
| resp = Run({"eval", "return redis.call('ZSCORE', KEYS[1], 'a') .. '-is-int'", "1", "z1"}); | |||
There was a problem hiding this comment.
I still need to understand. What did it return before?
There was a problem hiding this comment.
Specifically, maybe you can show the difference with tonumber command? I did not understand from your comment above where DF fails
dranikpg
force-pushed
the
fix-interpreter-double
branch
from
725c9ba to
25eb5aa
Compare
dranikpg
changed the title
dranikpg
force-pushed
the
fix-interpreter-double
branch
from
25eb5aa to
7359de0
Compare
Signed-off-by: Vladislav Oleshko <[email protected]>
dranikpg
force-pushed
the
fix-interpreter-double
branch
from
7359de0 to
7601c43
Compare
| static constexpr double kConvertEps = std::numeric_limits<double>::epsilon(); | ||
|
|
||
| double fractpart, intpart; | ||
| fractpart = modf(d, &intpart); |
There was a problem hiding this comment.
I'm not sure epsilon is correct if the fractional part is taken from a large number. Does it hold up to max long?
There was a problem hiding this comment.
They scale in the example for comparing the difference, https://en.cppreference.com/w/cpp/types/numeric_limits/epsilon and in my mind modf cant improve on accuracy if a large number is already stored skewed
But I checked in the range of int64-max and values far from power of twos and it seems that modf(d) < double_eps is always the case 🔦
No description provided.