Fix integer overflow in isapprox
#50730
Merged
Conversation
This file contains 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
nalimilan
commented
Jul 30, 2023
Comment on lines
+307
to
+309
| x′, y′ = promote(x, y) # to avoid integer overflow | ||
| x == y || | ||
| (isfinite(x) && isfinite(y) && norm(x-y) <= max(atol, rtol*max(norm(x′), norm(y′)))) || |
There was a problem hiding this comment.
Technically this can give slightly different results than before even in the absence of overflow, since promotion may round values larger than maxintfloat, giving a slightly different tolerance. But this should be OK.
| @@ -304,7 +304,20 @@ true | |||
| function isapprox(x::Number, y::Number; | |||
| atol::Real=0, rtol::Real=rtoldefault(x,y,atol), | |||
| nans::Bool=false, norm::Function=abs) | |||
| x == y || (isfinite(x) && isfinite(y) && norm(x-y) <= max(atol, rtol*max(norm(x), norm(y)))) || (nans && isnan(x) && isnan(y)) | |||
| x′, y′ = promote(x, y) # to avoid integer overflow | |||
There was a problem hiding this comment.
If we have the other method, why do we need to widen here?
There was a problem hiding this comment.
Because the other method is called only when both values are integer, but here norm(x) can overflow if x is an integer and y is a float. See tests added by the same commit.
EDIT: Note that the goal is not really to widen, but to switch to a floating point type so that instead of overflowing for values like typemax(Int8), norm gives a reasonable value to compute the tolerance.
stevengj
approved these changes
Aug 4, 2023
KristofferC
added
backport 1.9
LilithHafner
removed
the
merge me
KristofferC
pushed a commit
that referenced
this pull request
Aug 18, 2023
Ensure that `isapprox` gives correct results when comparing an integer with another integer or with a float. For comparison between integers, the fix only works when keeping default values for `rtol` and `norm`, and with `atol < 1`. It is not possible to handle the (atypical) case where `norm !== abs`, but that's OK since the user is responsible for providing a safe function. It would be possible to handle the case where `rtol > 0` or `atol >= 1`, but with complex code which would check for overflow and handle all possible corner cases; it would work only for types defined in Base and would not be extensible by packages. So I'm not sure that's worth it. At least with PR fixes the most common case. Fixes #50380. (cherry picked from commit 5f03a18)
IanButterworth
pushed a commit
that referenced
this pull request
Aug 19, 2023
Ensure that `isapprox` gives correct results when comparing an integer with another integer or with a float. For comparison between integers, the fix only works when keeping default values for `rtol` and `norm`, and with `atol < 1`. It is not possible to handle the (atypical) case where `norm !== abs`, but that's OK since the user is responsible for providing a safe function. It would be possible to handle the case where `rtol > 0` or `atol >= 1`, but with complex code which would check for overflow and handle all possible corner cases; it would work only for types defined in Base and would not be extensible by packages. So I'm not sure that's worth it. At least with PR fixes the most common case. Fixes #50380. (cherry picked from commit 5f03a18)
This was referenced Aug 19, 2023
IanButterworth
pushed a commit
that referenced
this pull request
Aug 25, 2023
Ensure that `isapprox` gives correct results when comparing an integer with another integer or with a float. For comparison between integers, the fix only works when keeping default values for `rtol` and `norm`, and with `atol < 1`. It is not possible to handle the (atypical) case where `norm !== abs`, but that's OK since the user is responsible for providing a safe function. It would be possible to handle the case where `rtol > 0` or `atol >= 1`, but with complex code which would check for overflow and handle all possible corner cases; it would work only for types defined in Base and would not be extensible by packages. So I'm not sure that's worth it. At least with PR fixes the most common case. Fixes #50380. (cherry picked from commit 5f03a18)
KristofferC
added a commit
that referenced
this pull request
Oct 2, 2023
Backported PRs: - [x] #48625 <!-- add replace(io, str, patterns...) --> - [x] #48387 <!-- Improve documentation of sort-related functions --> - [x] #48363 <!-- Revise sort.md and docstrings in sort.jl --> - [x] #48977 <!-- Update SparseArrays.jl stdlib for SuiteSparse 7 --> - [x] #50719 <!-- fix `CyclePadding(::DataType)` --> - [x] #50694 <!-- inference: permit recursive type traits --> - [x] #50860 <!-- Add `Base.get_extension` to docs/API --> - [x] #50594 <!-- Disallow non-index Integer types in isassigned --> - [x] #50802 <!-- Makes IntrusiveLinkedListSynchronized mutable to avoid allocation on poptask --> - [x] #50858 <!-- Add a `threadpool` parameter to `Channel` constructor --> - [x] #50874 <!-- Restrict COFF to a single thread when symbol count is high --> - [x] #50822 <!-- Add default method for setmodifiers! --> - [x] #50730 <!-- Fix integer overflow in `isapprox` --> - [x] #50850 <!-- Remove weird Rational dispatch and add pi functions to list --> - [x] #50809 <!-- Limit type-printing in MethodError --> - [x] #50915 <!-- Add note the `Task` about sticky bit --> - [x] #50929 <!-- when widening tuple types in tmerge, only widen the complex parts --> - [x] #50928 <!-- Bump JuliaSyntax to 0.4.6 --> - [x] #50959 <!-- Update libssh2 patches --> - [x] #50823 <!-- Make ranges more robust with unsigned indexes. --> - [x] #48542 <!-- Add docs on task-specific buffering using multithreading --> - [x] #50912 <!-- Separate foreign threads into a :foreign threadpool --> - [x] #51010 <!-- Add ORIGIN to SuiteSparse rpath on Linux/FreeBSD --> - [x] #50753 <!-- cat: remove unused promote_eltype methods that confuse inference --> - [x] #51027 <!-- Implement realloc accounting correctly --> - [x] #51019 <!-- fix a case of potentially use of undefined variable when handling error in distributed message processing --> - [x] #51039 <!-- Revert "Optimize findall(f, ::AbstractArray{Bool}) (#42202)" --> - [x] #51036 <!-- add missing invoke edge for nospecialize targets --> - [x] #51042 <!-- inference: fix return_type_tfunc modeling of concrete functions --> - [x] #51114 <!-- Workaround upstream FreeBSD issue #272992 --> - [x] #50892 <!-- Add `JL_DLLIMPORT` to `small_typeof` declaration --> - [x] #51154 <!-- broadcast: use recursion rather than ntuple to map over a tuple --> - [x] #51153 <!-- fix debug typo in "add missing invoke edge for nospecialize targets (#51036)" --> - [x] #51222 <!-- Check again if the tty is open inside the IO lock --> - [x] #51236 <!-- Add lock around uv_unref during init --> - [x] #51243 <!-- GMP: Gracefully handle more overflows. --> - [x] #51254 <!-- Ryu: make sure adding zeros does not overwrite trailing dot --> - [x] #51175 <!-- shorten stale_age for cachefile lock --> - [x] #51300 <!-- fix method definition error for bad vararg --> - [x] #51307 <!-- fix force-throw ctrl-C on Windows --> - [x] #51303 <!-- ensure revising structs is safe --> - [x] #51393 - [x] #51403 Need manual backport: - [x] #51009 <!-- fix #50562, regression in `in` of tuple of Symbols --> - [x] #51053 <!-- Bump Statistics stdlib --> - [x] #51013 <!-- fix #50709, type bound error due to free typevar in sparam env --> - [x] #51305 <!-- reduce test time for rounding and floatfuncs --> Contains multiple commits, manual intervention needed: - [ ] #50663 <!-- Fix Expr(:loopinfo) codegen --> - [ ] #51035 <!-- refactor GC scanning code to reflect jl_binding_t are now first class --> - [ ] #51092 <!-- inference: fix bad effects for recursion --> - [x] #51247 <!-- Check if malloc has succeeded before incrementing gc stats --> - [x] #51294 <!-- use LibGit2_jll for LibGit2 library --> Non-merged PRs with backport label: - [ ] #51132 <!-- Handle `AbstractQ` in concatenation --> - [x] #51029 <!-- testdefs: make sure that if a test set changes the active project, they change it back when they're done --> - [ ] #50919 <!-- Code loading: do the "skipping mtime check for stdlib" check regardless of the value of `ispath(f)` --> - [ ] #50824 <!-- Add some aliasing warnings to docstrings for mutating functions --> - [x] #50385 <!-- Precompile pidlocks: add to NEWS and docs --> - [ ] #49805 <!-- Limit TimeType subtraction to AbstractDateTime -->
KristofferC
removed
the
backport 1.10
nalimilan
added a commit
that referenced
this pull request
Nov 5, 2023
Ensure that `isapprox` gives correct results when comparing an integer with another integer or with a float. For comparison between integers, the fix only works when keeping default values for `rtol` and `norm`, and with `atol < 1`. It is not possible to handle the (atypical) case where `norm !== abs`, but that's OK since the user is responsible for providing a safe function. It would be possible to handle the case where `rtol > 0` or `atol >= 1`, but with complex code which would check for overflow and handle all possible corner cases; it would work only for types defined in Base and would not be extensible by packages. So I'm not sure that's worth it. At least with PR fixes the most common case. Fixes #50380. (cherry picked from commit 5f03a18)
KristofferC
added a commit
that referenced
this pull request
Nov 7, 2023
Backported PRs: - [x] #49357 <!-- Fix unclosed code fence in src/manual/methods.md --> - [x] #50842 <!-- Avoid race conditions with recursive rm --> - [x] #50858 <!-- Add a `threadpool` parameter to `Channel` constructor --> - [x] #50730 <!-- Fix integer overflow in `isapprox` --> - [x] #50823 <!-- Make ranges more robust with unsigned indexes. --> - [x] #50915 <!-- Add note the `Task` about sticky bit --> - [x] #50989 <!-- fix incorrect results in `expm1(::Union{Float16, Float32})` --> - [x] #50912 <!-- Separate foreign threads into a :foreign threadpool --> - [x] #51019 <!-- fix a case of potentially use of undefined variable when handling error in distributed message processing --> - [x] #51222 <!-- Check again if the tty is open inside the IO lock --> - [x] #51254 <!-- Ryu: make sure adding zeros does not overwrite trailing dot --> - [x] #51284 <!-- Avoid infinite loop when doing SIGTRAP in arm64-apple --> - [x] #51491 <!-- Throw clearer ArgumentError for strip with two string args --> - [x] #51531 <!-- fix `_tryonce_download_from_cache` (busybox.exe download error) -->
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Ensure that
isapproxgives correct results when comparing an integer with another integer or with a float. For comparison between integers, the fix only works when keeping default values forrtolandnorm, and withatol < 1.It is not possible to handle the (atypical) case where
norm !== abs, but that's OK since the user is responsible for providing a safe function.It would be possible to handle the case where
rtol > 0oratol >= 1, but with complex code which would check for overflow and handle all possible corner cases; it would work only for types defined in Base and would not be extensible by packages. So I'm not sure that's worth it. At least with PR fixes the most common case.Fixes #50380.