Quanlong Huang created ORC-314:
----------------------------------
Summary: C++ Reader crash for out of range scale in parsing
decimals
Key: ORC-314
URL: https://issues.apache.org/jira/browse/ORC-314
Project: ORC
Issue Type: Bug
Components: C++
Affects Versions: 1.4.3
Reporter: Quanlong Huang
Attachments: decimal_ill_scale.orc
The c++ reader crash in parsing the attached file. It's a corrupt orc file with
decimal values.
{code}
$ build/tools/src/orc-scan decimal_ill_scale.orc
Floating point exception (core dumped)
{code}
This is a devided-by-zero error:
{code}
Core was generated by `build/tools/src/orc-scan decimal_ill_scale.orc'.
Program terminated with signal SIGFPE, Arithmetic exception.
#0 0x0000000000896c0d in orc::Decimal64ColumnReader::readInt64
(this=0x1010120, value=@0x101aa60: 2222, currentScale=95) at
/mnt/volume1/orc/c++/src/ColumnReader.cc:1177
1177 value /= POWERS_OF_TEN[currentScale - scale];
(gdb) bt
#0 0x0000000000896c0d in orc::Decimal64ColumnReader::readInt64
(this=0x1010120, value=@0x101aa60: 2222, currentScale=95) at
/mnt/volume1/orc/c++/src/ColumnReader.cc:1177
#1 0x0000000000895700 in orc::Decimal64ColumnReader::next (this=0x1010120,
rowBatch=..., numValues=5, notNull=0x0) at
/mnt/volume1/orc/c++/src/ColumnReader.cc:1268
#2 0x0000000000893880 in orc::StructColumnReader::next (this=0xff9720,
rowBatch=..., numValues=5, notNull=0x0) at
/mnt/volume1/orc/c++/src/ColumnReader.cc:813
#3 0x0000000000833e86 in orc::RowReaderImpl::next (this=0x10044b0, data=...)
at /mnt/volume1/orc/c++/src/Reader.cc:832
#4 0x000000000082d274 in scanFile (out=..., filename=0x7ffc132d3755
"decimal_ill_scale.orc", batchSize=1024) at
/mnt/volume1/orc/tools/src/FileScan.cc:39
#5 0x000000000082d4c5 in main (argc=1, argv=0x7ffc132d1940) at
/mnt/volume1/orc/tools/src/FileScan.cc:84
(gdb) p currentScale
$1 = 95
(gdb) p scale
$2 = 0
{code}
95 is out of the length of the POWERS_OF_TEN array. So it gets value 0
unexpectedly.
{code}
class Decimal64ColumnReader: public ColumnReader {
public:
static const uint32_t MAX_PRECISION_64 = 18;
static const uint32_t MAX_PRECISION_128 = 38;
static const int64_t POWERS_OF_TEN[MAX_PRECISION_64 + 1];
{code}
We should check the range of currentScale in Decimal64ColumnReader::readInt64
to avoid this.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
