[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 Sakura286 changed: What|Removed |Added Status|NEW |ASSIGNED --- Comment #12 from Sakura286 --- Here is the patch which add NaN payload check at app startup. https://gerrit.libreoffice.org/c/core/+/165391 The content of the patch: 1. check nan payload on app startup 2. emit signal to call 'CreateNanPayloadWarningDialog()' 3. the factory creates dialog 4. record property 'ShowNanPayloadWarning' persistent -- You are receiving this mail because: You are the assignee for the bug.
[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #11 from Eike Rathke --- #3 is not possible as there is no non-NaN-payload mode implemented, and implementing it would require not only different error handling but also different transport of errors between intermediary results up to final results of a formula expression; while technically possible I wouldn't call it feasible or viable, as the effort would be huge, add even more runtime penalty, and possibly overlooking things adds uncertainty. #2 might work to some degree, someone would have to actually try on such an inferior architecture what happens. My (hopeful) guess is that _some_ NaN is transported so there will be _some_ error (e.g. indicating numeric failure) but any detail about what specific error will be lost. But that also means that code reacting on specific error information will fail, so not everything will work. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #10 from Sakura286 --- I prefer #2, too. I am wondering whether #2 is acceptable by community. If so, I am willing to do a bit of work on this. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #9 from Rene Engelhard --- I'd personally prefer 2. 1. has the problem that one still ships libreoffice-calc. Shipping something which doesn't work (fails on start, even with a message) is not good. Or 3, but I don't think there's such a mode (yet). And if we had such a mode we wouldn't need that NaN payload thingy at all. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #8 from Sakura286 --- (In reply to Eike Rathke from comment #5) > We might have to have a runtime check in Calc as well and bail out if NaN > payloads are not supported. Yes, a compile time check cannot guarantee that this functionality works well at runtime. So what behaviour should Calc take when detects the lack of NaN payload support at runtime? As I understand it, there might be three solutions: 1. Giving a warning and shutting down calc. 2. Giving a warning and disclaimer. Calc continues running. 3. Silently switch to a non-NaN payload mode. For #1, maybe it is a bit distructive to directly rule out RISC-V as a supported platform. #2 is a transitional solution. For #3, is it possible to add this feature? According to my observation, on riscv64, Calc does return error code when a formula error occurs, but sometimes it returns wrong error code '#NUM'. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #7 from Eike Rathke --- No explicit documents, but Calc heavily uses quiet NaNs to transport and propagate any runtime error in the calculation interpreter engine and relies on NaN payload propagation in all math operators. See CreateDoubleError() and GetDoubleErrorValue() at https://opengrok.libreoffice.org/xref/core/include/formula/errorcodes.hxx?r=807f238f#96 and their use https://opengrok.libreoffice.org/s?refs=CreateDoubleError=core https://opengrok.libreoffice.org/s?refs=GetDoubleErrorValue=core -- You are receiving this mail because: You are the assignee for the bug.
[Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #6 from Chen Xuan --- Currently NaN payload are not supported by most RISC-V machines. It might be necessary for RISC-V porters to know which component is affected by NaN payload. Are there any documents that describe how NaN payload work in LibreOffice? -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 BogdanB changed: What|Removed |Added Blocks||108827 CC||buzea.bog...@libreoffice.or ||g Referenced Bugs: https://bugs.documentfoundation.org/show_bug.cgi?id=108827 [Bug 108827] [META] Calc functions bugs and enhancements -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #5 from Eike Rathke --- From https://groups.google.com/a/groups.riscv.org/g/isa-dev/c/g79dHlV4B_k/m/dtlPNaq8AwAJ "Implementors are free to provide a NaN payload propagation scheme as a nonstandard extension enabled by a nonstandard operating mode." I deduce that there may be RISC-V systems that support propagating NaN payloads. So with bad luck a build, even if it did make check on such system, would not run properly on systems that do not support the extension. We might have to have a runtime check in Calc as well and bail out if NaN payloads are not supported. -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 Eike Rathke changed: What|Removed |Added Status|UNCONFIRMED |NEW Ever confirmed|0 |1 Version|7.6.0.0 alpha0+ Master |Inherited From OOo OS|Linux (All) |All -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 --- Comment #4 from Commit Notification --- Eike Rathke committed a patch related to this issue. It has been pushed to "master": https://git.libreoffice.org/core/commit/2fd915b7d2d2a9e25600a07587ab01d23bffbc58 Related: tdf#152943 Test that a quiet NaN payload is propagated It will be available in 7.6.0. The patch should be included in the daily builds available at https://dev-builds.libreoffice.org/daily/ in the next 24-48 hours. More information about daily builds can be found at: https://wiki.documentfoundation.org/Testing_Daily_Builds Affected users are encouraged to test the fix and report feedback. -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 Commit Notification changed: What|Removed |Added Whiteboard||target:7.6.0 -- You are receiving this mail because: You are the assignee for the bug.
[Libreoffice-bugs] [Bug 152943] GetErrCode() doesn't work as expected on RISC-V due to floating-point NaN payload not being propagated
https://bugs.documentfoundation.org/show_bug.cgi?id=152943 Eike Rathke changed: What|Removed |Added Summary|GetErrCode() doesn't work |GetErrCode() doesn't work |as expected on RISC-V due |as expected on RISC-V due |to invalid floating-point |to floating-point NaN |operation in KahanSum |payload not being ||propagated -- You are receiving this mail because: You are the assignee for the bug.
