Fix formula for new_size in WriteBufferFromVector(AppendModeTag). #6208
Conversation
dbms/src/IO/WriteBuffer.h
Outdated
| @@ -35,8 +35,6 @@ class WriteBuffer : public BufferBase | |||
| */ | |||
| inline void next() | |||
| { | |||
| if (!offset()) | |||
| return; | |||
There was a problem hiding this comment.
It seems WriteBuffer::nextIfAtEnd() should only check that the cursor reaches the end of the buffer before calling WriteBuffer::nextImpl().
There was a problem hiding this comment.
Will it lead to writing zero bytes into file; compressing block of data with zero size, etc?
There was a problem hiding this comment.
I think you're right and it's better not to leave a WriteBuffer with empty size after initialization. So I've removed my changes in WriteBuffer.h and just corrected formula in WriteBufferFromVector(AppendModeTag).
dbms/src/IO/WriteBufferFromVector.h
Outdated
| @@ -57,7 +57,7 @@ class WriteBufferFromVector : public WriteBuffer | |||
| : WriteBuffer(nullptr, 0), vector(vector_) | |||
| { | |||
| size_t old_size = vector.size(); | |||
| vector.resize(std::max(vector.size() + initial_size, vector.capacity())); | |||
| vector.resize(std::max(initial_size, vector.capacity())); | |||
There was a problem hiding this comment.
The change in WriteBuffer::next() fixes the hanging. We can still add initial_size to vector.size() here but it seems it's better not to do it in this constructor in advance because WriteBuffer::nextImpl() also can increase the size of the vector.
There was a problem hiding this comment.
This commend isn't actual anymore. Here I only corrected the formula for consistency with WriteBufferFromVector::nextImpl().
vitlibar
force-pushed
the
write-buffer-call-next-impl-even-if-zero-offset
branch
from
870c7d0
to
b51898f
Compare
vitlibar
changed the title
vitlibar
force-pushed
the
write-buffer-call-next-impl-even-if-zero-offset
branch
from
b51898f
to
d7891ec
Compare
KochetovNicolai
added
the
pr-bugfix
…en-if-zero-offset Fix formula for new_size in WriteBufferFromVector(AppendModeTag). (cherry picked from commit e9c2077)
…en-if-zero-offset Fix formula for new_size in WriteBufferFromVector(AppendModeTag). (cherry picked from commit e9c2077)
I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en
Category (leave one):
It's another approach to fix #6195