Re: [Patch, fortran] PR117797 - [13/14/15 Regression] ICE in gfc_get_array_span
Hi Paul, no, I don't have an example in mind out of the box. May be something like: type :: T character :: s(:) character :: c end type class(T) :: foo(2) But that is just a guess. I was triggered only by the first condition your patch removes, where span was used instead of the of element-size. I was more or less just curious. Let's see what time brings to our regression shores. - Andre On Tue, 10 Dec 2024 17:45:09 + Paul Richard Thomas wrote: > Hi Andre, > > I don't think that's possible but I am prepared to be proven wrong :-) Do > you have an example in mind, where this might happen? > > Regards > > Paul > > > On Tue, 10 Dec 2024 at 10:52, Andre Vehreschild wrote: > > > Hi Paul, > > > > in principle the patch looks sound, but is the vptr_size always the same > > as a > > descriptor->span? Can't the later be larger because of padding? If that is > > the > > case, then the first `else if (UNLIMITED...` you removed in > > `gfc_get_array_span > > ()` would return a larger number. > > > > Regards, > > Andre > > > > On Tue, 10 Dec 2024 10:10:07 + > > Paul Richard Thomas wrote: > > > > > Hi All, > > > > > > This was yet another regression that I caused, which was backported and > > so > > > I am rather anxious to fix it promptly. > > > > > > The modifications that I made to gfc_get_array_span caused > > > unlimited polymorphic array components to be missed, when contained in a > > > dummy. Instead, the dummy was taken to be the class container, which then > > > caused the segfault > > > > > > I took the opportunity to refactor gfc_get_array_span such that it is now > > > improved over its original form. Class data descriptors now cannot be > > > missed for any declared type. The new helper function > > > 'class_array_element_size' extracts the vptr from both the class object > > and > > > from class dummies. The vptr size is then returned, except in the case of > > > unlimited polymorphic expressions, where the vptr size must be corrected > > > with the _len value. The class container is unambiguously obtainable from > > > the vptr. > > > > > > Regression tests - OK for mainline and then, after a short while, > > > backported to the affected branches? > > > > > > Paul > > > > > > -- > > Andre Vehreschild * Email: vehre ad gmx dot de > > -- Andre Vehreschild * Email: vehre ad gmx dot de
Re: [Patch, fortran] PR117797 - [13/14/15 Regression] ICE in gfc_get_array_span
Hi Andre, I don't think that's possible but I am prepared to be proven wrong :-) Do you have an example in mind, where this might happen? Regards Paul On Tue, 10 Dec 2024 at 10:52, Andre Vehreschild wrote: > Hi Paul, > > in principle the patch looks sound, but is the vptr_size always the same > as a > descriptor->span? Can't the later be larger because of padding? If that is > the > case, then the first `else if (UNLIMITED...` you removed in > `gfc_get_array_span > ()` would return a larger number. > > Regards, > Andre > > On Tue, 10 Dec 2024 10:10:07 + > Paul Richard Thomas wrote: > > > Hi All, > > > > This was yet another regression that I caused, which was backported and > so > > I am rather anxious to fix it promptly. > > > > The modifications that I made to gfc_get_array_span caused > > unlimited polymorphic array components to be missed, when contained in a > > dummy. Instead, the dummy was taken to be the class container, which then > > caused the segfault > > > > I took the opportunity to refactor gfc_get_array_span such that it is now > > improved over its original form. Class data descriptors now cannot be > > missed for any declared type. The new helper function > > 'class_array_element_size' extracts the vptr from both the class object > and > > from class dummies. The vptr size is then returned, except in the case of > > unlimited polymorphic expressions, where the vptr size must be corrected > > with the _len value. The class container is unambiguously obtainable from > > the vptr. > > > > Regression tests - OK for mainline and then, after a short while, > > backported to the affected branches? > > > > Paul > > > -- > Andre Vehreschild * Email: vehre ad gmx dot de >
Re: [Patch, fortran] PR117797 - [13/14/15 Regression] ICE in gfc_get_array_span
Hi Paul, in principle the patch looks sound, but is the vptr_size always the same as a descriptor->span? Can't the later be larger because of padding? If that is the case, then the first `else if (UNLIMITED...` you removed in `gfc_get_array_span ()` would return a larger number. Regards, Andre On Tue, 10 Dec 2024 10:10:07 + Paul Richard Thomas wrote: > Hi All, > > This was yet another regression that I caused, which was backported and so > I am rather anxious to fix it promptly. > > The modifications that I made to gfc_get_array_span caused > unlimited polymorphic array components to be missed, when contained in a > dummy. Instead, the dummy was taken to be the class container, which then > caused the segfault > > I took the opportunity to refactor gfc_get_array_span such that it is now > improved over its original form. Class data descriptors now cannot be > missed for any declared type. The new helper function > 'class_array_element_size' extracts the vptr from both the class object and > from class dummies. The vptr size is then returned, except in the case of > unlimited polymorphic expressions, where the vptr size must be corrected > with the _len value. The class container is unambiguously obtainable from > the vptr. > > Regression tests - OK for mainline and then, after a short while, > backported to the affected branches? > > Paul -- Andre Vehreschild * Email: vehre ad gmx dot de
