Skip to content

Using the public PyUnicodeWriter C API made the json module slower #133968

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
vstinner opened this issue May 13, 2025 · 5 comments
Closed

Using the public PyUnicodeWriter C API made the json module slower #133968

vstinner opened this issue May 13, 2025 · 5 comments
Labels
extension-modules C modules in the Modules dir performance Performance or resource usage stdlib Python modules in the Lib dir type-bug An unexpected behavior, bug, or error

Comments

@vstinner
Copy link
Member

vstinner commented May 13, 2025 •
edited by bedevere-app bot
Loading

I modified the json module to replace the private _PyUnicodeWriter C API with the public PyUnicodeWriter C API:

Problem: it made the json module slower. Let's investigate what's going on.

Linked PRs
@vstinner vstinner added performance Performance or resource usage stdlib Python modules in the Lib dir labels May 13, 2025
@bedevere-app bedevere-app bot mentioned this issue May 13, 2025
Merged
@vstinner
Copy link
Member Author

See also #133186.

vstinner added a commit to vstinner/cpython that referenced this issue May 13, 2025
@vstinner
pythongh-133968: Add fast path to PyUnicodeWriter_WriteStr()
8613170 
Don't call PyObject_Str() if the input type is str.
@bedevere-app bedevere-app bot mentioned this issue May 13, 2025
Merged
vstinner added a commit that referenced this issue May 13, 2025
@vstinner
gh-133968: Add fast path to PyUnicodeWriter_WriteStr() (#133969)
fe9f6e8 
Don't call PyObject_Str() if the input type is str.
miss-islington pushed a commit to miss-islington/cpython that referenced this issue May 13, 2025
@vstinner @miss-islington
pythongh-133968: Add fast path to PyUnicodeWriter_WriteStr() (pythonG…
616ee64 
…H-133969)

Don't call PyObject_Str() if the input type is str.
(cherry picked from commit fe9f6e8)

Co-authored-by: Victor Stinner <vstinner@python.org>
@bedevere-app bedevere-app bot mentioned this issue May 13, 2025
Merged
vstinner added a commit that referenced this issue May 13, 2025
@miss-islington @vstinner
[3.14] gh-133968: Add fast path to PyUnicodeWriter_WriteStr() (GH-133969
5632e93 
) (#133971)

gh-133968: Add fast path to PyUnicodeWriter_WriteStr() (GH-133969)

Don't call PyObject_Str() if the input type is str.
(cherry picked from commit fe9f6e8)

Co-authored-by: Victor Stinner <vstinner@python.org>
vstinner added a commit to vstinner/cpython that referenced this issue May 13, 2025
@vstinner
pythongh-133968: Add PyUnicodeWriter_WriteASCII() function
fa9d9df 
Replace most PyUnicodeWriter_WriteUTF8() calls with
PyUnicodeWriter_WriteASCII().
vstinner added a commit to vstinner/cpython that referenced this issue May 13, 2025
@vstinner
pythongh-133968: Add PyUnicodeWriter_WriteASCII() function
2aa2e87 
Replace most PyUnicodeWriter_WriteUTF8() calls with
PyUnicodeWriter_WriteASCII().
@bedevere-app bedevere-app bot mentioned this issue May 13, 2025
Merged
@vstinner vstinner mentioned this issue May 14, 2025
Closed
6 tasks
@encukou
Copy link
Member

encukou commented May 15, 2025

Could you try overoptimizing PyUnicodeWriter_WriteUTF8 and benchmarking that, to see how fast the existing function can “theoretically” be?
I'm thinking something like:

  • in the JSON module, force the constants ("null" etc) to be size_t-aligned
  • add private (ABI-only) function _PyUnicodeWriter_WriteUTF8_SmallAligned, which requires size_t-aligned input and size <= sizeof(size_t)
  • make PyUnicodeWriter_WriteUTF8 a macro that calls _PyUnicodeWriter_WriteUTF8_SmallAligned if the need are met, and PyUnicodeWriter_WriteUTF8 otherwise

@picnixz picnixz added extension-modules C modules in the Modules dir type-bug An unexpected behavior, bug, or error labels May 15, 2025
@vstinner
Copy link
Member Author

in the JSON module, force the constants ("null" etc) to be size_t-aligned

I don't know how to guarantee that.

add private (ABI-only) function _PyUnicodeWriter_WriteUTF8_SmallAligned, which requires size_t-aligned input and size <= sizeof(size_t)

Why do you want to limit the size to sizeof(size_t)? Which kind of optimization are you thinking of? find_first_nonascii() already takes care of the different cases with different optimizations.

ascii_decode() is already highly optimized.

@encukou
Copy link
Member

encukou commented May 17, 2025

ascii_decode() is already highly optimized.

Indeed it is! After playing around a bit, I could only optimize short-ASCII-strings at the expense of other strings, or a lot of code duplication.
Sorry for the noise!

I don't know how to guarantee that.

For the record: alignas; in old C it could be union { char my_string[N]; size_t _align_me_please; }

vstinner added a commit that referenced this issue May 29, 2025
@vstinner @ZeroIntensity @picnixz
gh-133968: Add PyUnicodeWriter_WriteASCII() function (#133973)
f49a07b 
Replace most PyUnicodeWriter_WriteUTF8() calls with
PyUnicodeWriter_WriteASCII().

Unrelated change to please the linter: remove an unused
import in test_ctypes.

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
vstinner added a commit to vstinner/cpython that referenced this issue May 31, 2025
@vstinner @ZeroIntensity @picnixz
pythongh-133968: Add PyUnicodeWriter_WriteASCII() function (python#13…
1dc045c 
…3973)

Replace most PyUnicodeWriter_WriteUTF8() calls with
PyUnicodeWriter_WriteASCII().

Unrelated change to please the linter: remove an unused
import in test_ctypes.

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
(cherry picked from commit f49a07b)
@bedevere-app bedevere-app bot mentioned this issue May 31, 2025
Merged
vstinner pushed a commit that referenced this issue May 31, 2025
@nineteendo
gh-133968: Create the Unicode writer on demand in json (#133832)
c81446a
vstinner added a commit to vstinner/cpython that referenced this issue Jun 2, 2025
@vstinner @ZeroIntensity @picnixz
pythongh-133968: Add PyUnicodeWriter_WriteASCII() function (python#13…
39506d1 
…3973)

Replace most PyUnicodeWriter_WriteUTF8() calls with
PyUnicodeWriter_WriteASCII().

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
(cherry picked from commit f49a07b)
vstinner added a commit that referenced this issue Jun 9, 2025
@vstinner @ZeroIntensity @picnixz
[3.14] gh-133968: Add PyUnicodeWriter_WriteASCII() function (#133973) (…
3d69d18 
…#134974)

gh-133968: Add PyUnicodeWriter_WriteASCII() function (#133973)

Replace most PyUnicodeWriter_WriteUTF8() calls with
PyUnicodeWriter_WriteASCII().

(cherry picked from commit f49a07b)

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
vstinner added a commit to vstinner/cpython that referenced this issue Jun 9, 2025
@vstinner
pythongh-133968: Update PyUnicodeWriter_WriteASCII() documentation
8186f37 
The function was added to Python 3.14.
@bedevere-app bedevere-app bot mentioned this issue Jun 9, 2025
Merged
vstinner added a commit that referenced this issue Jun 9, 2025
@vstinner
gh-133968: Update PyUnicodeWriter_WriteASCII() documentation (#135297)
aaad2e8 
The function was added to Python 3.14.
@vstinner
Copy link
Member Author

vstinner commented Jun 9, 2025

I close the issue. Multiple changes optimized the json module which should be "less slow" than 3.13 in 3.14 and main (3.15) branches now.

If someone has more optimization ideas, I suggest opening a new issue.

@vstinner vstinner closed this as completed Jun 9, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
extension-modules C modules in the Modules dir performance Performance or resource usage stdlib Python modules in the Lib dir type-bug An unexpected behavior, bug, or error
Projects
JSON issues
Status: Done
Milestone
No milestone
Development

No branches or pull requests
3 participants
@vstinner @encukou @picnixz