abhinav-phi opened a new pull request, #68733:
URL: https://github.com/apache/doris/pull/68733
### What problem does this PR solve?
Issue Number: close #68722
Related PR: #68729 (same file, different method — that one is the BOOLEAN
branch, this is the
integral branch; whichever lands first the other rebases without conflict)
Problem Summary:
`FractionalLiteral.uncheckedCastTo(<integral type>)` built its working
decimal with
`new BigDecimal(value.toString())`. For a `Float`/`Double` that is the
*shortest decimal
representation* of the value, not the value itself, so the folded integral
is derived from the
text rather than from the binary magnitude. BE converts the binary value.
```
SELECT CAST(CAST(1e38 AS DOUBLE) AS LARGEINT) AS fe, CAST(d1 AS LARGEINT) AS
be FROM t;
+-----------------------------------------+------------------------------------------+
| 100000000000000000000000000000000000000 |
99999999999999997748809823456034029568 |
+-----------------------------------------+------------------------------------------+
```
`1e38` as a double really is `99999999999999997748809823456034029568`;
`Double.toString(1e38)` is
`"1.0E38"`, and `new BigDecimal("1.0E38")` is exactly 10^38 — hence the
mismatch, and a constant
fold that LARGEINT cannot even represent round-trip.
The fix keeps the decimal-string behaviour for `BigDecimal` sources (decimal
literals are already
exact) and uses the binary value for `Float`/`Double` sources, which is what
BE does. Truncation
toward zero, the finite check and the overflow check are unchanged, so
`CAST(12.999 AS INT) = 12`, `CAST(-12.999 AS BIGINT) = -12` and out-of-range
values still throw
`CastException` — all of which the existing assertions still cover.
The date-like branch of the same method still uses `getValue().toString()`;
a `DOUBLE` -> `DATE`
fold goes through a day-number encoding where the decimal-vs-binary question
needs its own look at
BE, so it is deliberately left out of this change.
### Release note
None
### Check List (For Author)
- Test <!-- At least one of them must be included. -->
- [ ] Regression test
- [x] Unit Test
- [ ] Manual test (add detailed scripts or steps below)
- [ ] No need to test or manual test. Explain why:
New `DoubleLiteralTest.testCastToIntegralUsesTheBinaryValue` asserts the
reported case
(`CAST(1e38 AS LARGEINT)` -> `99999999999999997748809823456034029568`),
that the exactly
representable case is unchanged (`CAST(1e9 AS LARGEINT)` -> `1000000000`),
that ordinary
truncation still trims toward zero (`CAST(-12.999 AS BIGINT)` -> `-12`)
and that an out-of-range
magnitude still raises `CastException` instead of saturating.
Red, on `master` with only the test added:
```
[ERROR] DoubleLiteralTest.testCastToIntegralUsesTheBinaryValue:388
expected: <99999999999999997748809823456034029568> but was:
<100000000000000000000000000000000000000>
```
Green, with the fix (whole net):
```
DoubleLiteralTest Tests run: 3, Failures: 0 (focused run: BUILD
SUCCESS, exit 0)
FloatLiteralTest Tests run: 3, Failures: 0
LargeIntLiteralTest Tests run: 3, Failures: 0
IntegerLiteralTest Tests run: 2, Failures: 0
DecimalLiteralTest Tests run: 2, Failures: 0
CompareLiteralTest Tests run: 2, Failures: 0
CastTest Tests run: 23, Failures: 0
ConstantProjectionFoldingTest Tests run: 1, Failures: 0
org.apache.doris.analysis.FloatLiteralTest Tests run: 2, Failures: 0
```
Two `FoldConstantTest` failures remain in the net; they are **pre-existing
on this machine and
unrelated** — they reproduce identically on pristine `master` with every
edit of mine stashed, and
they are session-time-zone dependent timestamp assertions:
```
[ERROR] FoldConstantTest.testDateConstructFunction:1552 expected:
<'1977-06-03 17:57:24'> but was: <'1977-06-03 15:27:24'>
[ERROR] FoldConstantTest.testFoldString:581 expected: <'2025-10-27
14:58:08.100000'> but was: <'2025-10-27 12:28:08.100000'>
```
Command (project flags from `run-fe-ut.sh`):
```
mvn test -pl fe-core -am -Dcheckstyle.skip=true -DfailIfNoTests=false \
-Dmaven.build.cache.enabled=false -Dtest=DoubleLiteralTest
```
Environment: JDK 17.0.20 (Temurin), Maven 3.9.16, thrift 0.24.0 from the
prebuilt thirdparty
package, Ubuntu 22.04 under WSL2. Style gate `mvn -pl fe-core
checkstyle:check` ->
`You have 0 Checkstyle violations.`
- Behavior changed:
- [ ] No.
- [x] Yes. <!-- Explain the behavior change -->
Folding a `FLOAT`/`DOUBLE` literal to an integral type now uses the
binary value, so cases
where the shortest decimal differs from the binary magnitude (the
reported `1e38` -> LARGEINT)
return the BE result. Values that are exactly representable, and all
truncation/overflow
behaviour, are unchanged.
- Does this need documentation?
- [x] No.
- [ ] Yes. <!-- Add document PR link here. -->
### Check List (For Reviewer who merge this PR)
- [ ] Confirm the release note
- [ ] Confirm test cases
- [ ] Confirm document
- [ ] Add branch pick label
### AI assistance disclosure
An AI coding agent was used to draft this change and the new test. I read
and reviewed every line
myself, ran the FE unit tests above on the exact commit in this branch
(including the pristine-master
control run for `FoldConstantTest`), and I take responsibility for the
result.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
