Closed
Conversation
(pruning). We tag parsed tokens to associate a branch identifier to them
…atibility reasons
|
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. |
frantzmiccoli
force-pushed
the
feature/branch-pruning
branch
2 times, most recently
from
aa38599 to
c7e3e8b
Compare
Contributor
Author
|
Did a merge with master to avoid merge conflicts |
frantzmiccoli
force-pushed
the
feature/branch-pruning
branch
from
c7e3e8b to
da19445
Compare
frantzmiccoli
force-pushed
the
feature/branch-pruning
branch
from
da19445 to
4d8ca2a
Compare
Member
|
@MarkBaker this PR still look quite promising should we merge it now ? or would it conflict with your work on the new calculation engine ? |
Contributor
Author
|
Side note: we have been using in production since a few weeks now on some pretty advanced sheet without observing any bugs |
Member
|
Thank you for your work and patience. I finally squashed and merged it as 0b387e7 |
PowerKiKi
added a commit
that referenced
this pull request
Aug 17, 2019
1.9.0 ### Added - When <br> appears in a table cell, set the cell to wrap [#1071](#1071) and [#1070](#1070) - Add MAXIFS, MINIFS, COUNTIFS and Remove MINIF, MAXIF [#1056](#1056) - HLookup needs an ordered list even if range_lookup is set to false [#1055](#1055) and [#1076](#1076) - Improve performance of IF function calls via ranch pruning to avoid resolution of every branches [#844](#844) - MATCH function supports `*?~` Excel functionality, when match_type=0 [#1116](#1116) - Allow HTML Reader to accept HTML as a string [#1136](#1136) ### Fixed - Fix to AVERAGEIF() function when called with a third argument - Eliminate duplicate fill none style entries [#1066](#1066) - Fix number format masks containing literal (non-decimal point) dots [#1079](#1079) - Fix number format masks containing named colours that were being misinterpreted as date formats; and add support for masks that fully replace the value with a full text string [#1009](#1009) - Stricter-typed comparison testing in COUNTIF() and COUNTIFS() evaluation [#1046](#1046) - COUPNUM should not return zero when settlement is in the last period [#1020](#1020) and [#1021](#1021) - Fix handling of named ranges referencing sheets with spaces or "!" in their title - Cover `getSheetByName()` with tests for name with quote and spaces [#739](#739) - Best effort to support invalid colspan values in HTML reader - [#878](#878) - Fixes incorrect rows deletion [#868](#868) - MATCH function fix (value search by type, stop search when match_type=-1 and unordered element encountered) [#1116](#1116) - Fix `getCalculatedValue()` error with more than two INDIRECT [#1115](#1115) - Writer\Html did not hide columns [#985](#985)
1 task
BlackyTay
pushed a commit
to BlackyTay/PhpSpreadsheet
that referenced
this pull request
Aug 8, 2025
…branches Calculation engine was resolving every function by first resolving its arguments including IFs, this was causing significant over evaluation when IFs were used as it meant for every case to be evaluated. Introduce elements to identify ifs and enable better branch resolution (pruning). We tag parsed tokens to associate a branch identifier to them. Closes PHPOffice#844
BlackyTay
pushed a commit
to BlackyTay/PhpSpreadsheet
that referenced
this pull request
Aug 8, 2025
1.9.0 ### Added - When <br> appears in a table cell, set the cell to wrap [PHPOffice#1071](PHPOffice#1071) and [PHPOffice#1070](PHPOffice#1070) - Add MAXIFS, MINIFS, COUNTIFS and Remove MINIF, MAXIF [PHPOffice#1056](PHPOffice#1056) - HLookup needs an ordered list even if range_lookup is set to false [PHPOffice#1055](PHPOffice#1055) and [PHPOffice#1076](PHPOffice#1076) - Improve performance of IF function calls via ranch pruning to avoid resolution of every branches [PHPOffice#844](PHPOffice#844) - MATCH function supports `*?~` Excel functionality, when match_type=0 [PHPOffice#1116](PHPOffice#1116) - Allow HTML Reader to accept HTML as a string [PHPOffice#1136](PHPOffice#1136) ### Fixed - Fix to AVERAGEIF() function when called with a third argument - Eliminate duplicate fill none style entries [PHPOffice#1066](PHPOffice#1066) - Fix number format masks containing literal (non-decimal point) dots [PHPOffice#1079](PHPOffice#1079) - Fix number format masks containing named colours that were being misinterpreted as date formats; and add support for masks that fully replace the value with a full text string [PHPOffice#1009](PHPOffice#1009) - Stricter-typed comparison testing in COUNTIF() and COUNTIFS() evaluation [PHPOffice#1046](PHPOffice#1046) - COUPNUM should not return zero when settlement is in the last period [PHPOffice#1020](PHPOffice#1020) and [PHPOffice#1021](PHPOffice#1021) - Fix handling of named ranges referencing sheets with spaces or "!" in their title - Cover `getSheetByName()` with tests for name with quote and spaces [PHPOffice#739](PHPOffice#739) - Best effort to support invalid colspan values in HTML reader - [PHPOffice#878](PHPOffice#878) - Fixes incorrect rows deletion [PHPOffice#868](PHPOffice#868) - MATCH function fix (value search by type, stop search when match_type=-1 and unordered element encountered) [PHPOffice#1116](PHPOffice#1116) - Fix `getCalculatedValue()` error with more than two INDIRECT [PHPOffice#1115](PHPOffice#1115) - Writer\Html did not hide columns [PHPOffice#985](PHPOffice#985)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Repost of #818 after a rebase on
master.This is:
Checklist:
Why this change is needed?
Calculation engine was resolving every function by first resolving its arguments including IFs, this was causing significant over evaluation when IFs were used as it meant for every case to be evaluated.
I have tested against 5 files made by 4 different people to ensure that this was not introducing regression, I have observed none. It generates speed improvement from 0% to 80% on those files.
EDIT Completing from the discussion I had with @PowerKiKi
Yes, this is touching the core of the calculation engine, so, it would not be to surprising to introduce regression. I did my best to thoroughly test it, if you see extra tests to perform let me know.
As per the extra public methods:
Stack::getStackItem(): enables code factorization asCalculation::_parseFormula()doesn't manipulate tokens only through a stack and was also manually creating the arrays representing tokens.Stack::__toString(): was really convenient to debug I would recommend to leave it but I could truncate it.Calculation:setBranchPruningEnabled(),enableBranchPruning(),disableBranchPruning()andclearBranchStore().processTokenStack(), I usually prefer to use the reflection API in my unit tests.A side point about
CalculationTest::testBranchPruningFormulaParsing, this test was pretty thick. I could not use a data provider as the expected result testing is way non trivial. So I did split the test in many different functions.As per the code splitting in the calculation engine itself, the
_parseFormula()andprocessTokenStack()functions are not using context objects that could be pass around to subrountines. I could spend some time splitting those huge methods as I think I have gained some understanding of the calculation engine inner workings but I think this should be for another pull request. I don't think it would deeply impact execution speed.