Bug 150403

Summary: JSC Math.pow tests makes wrong assumptions on precision
Product: WebKit Reporter: Guillaume Emont <guijemont>
Component: JavaScriptCoreAssignee: Nobody <webkit-unassigned>
Status: RESOLVED INVALID    
Severity: Normal CC: fpizlo
Priority: P2    
Version: WebKit Nightly Build   
Hardware: Unspecified   
OS: Unspecified   
Bug Depends on:    
Bug Blocks: 108664    
Attachments:
Description Flags
Patch
none
Patch fpizlo: review-

Guillaume Emont
Reported 2015-10-21 11:42:44 PDT
The test in Source/JavaScriptCore/tests/stress/math-pow-with-constants.js assumes the results of Math.pow() calls would be of a given precision, even though the spec[1] says that Math.pow(): Returns an implementation-dependent approximation to the result of raising x to the power y. At least on MIPS, the precision of the result is different on some tests. [1] http://www.ecma-international.org/ecma-262/6.0/#sec-math.pow
Attachments
Patch (2.60 KB, patch)
2015-10-21 11:50 PDT, Guillaume Emont 		no flags
DetailsFormatted DiffDiff
Patch (2.57 KB, patch)
2015-10-21 15:43 PDT, Guillaume Emont 		fpizlo: review-
DetailsFormatted DiffDiff
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.
Guillaume Emont
Comment 1 2015-10-21 11:50:20 PDT
Created attachment 263708 [details] Patch
Guillaume Emont
Comment 2 2015-10-21 15:43:54 PDT
Created attachment 263752 [details] Patch New version with typo corrected
Filip Pizlo
Comment 3 2015-10-21 19:47:31 PDT
Comment on attachment 263752 [details] Patch Sounds like a mips bug. Jsc aims to have strong precision guarantees.
Guillaume Emont
Comment 4 2015-10-22 11:30:51 PDT
After further checking, it seems like the issue that I encountered with that test only happens when using uclibc. In that case, Math.pow(2,42.5) returns 6219777023950.949. Though that number looks like it has more precision, it turns out[1] that it seems that it is not rounded properly, but rather truncated (the correct rounding to 3 digits would be 6219777023950.950), so it seems like it is a uclibc issue. Sorry for the noise. [1] http://www.wolframalpha.com/input/?i=2%5E42.5
Note You need to log in before you can comment on or make changes to this bug.