diff options
author | Bartosz Kosiorek <gang65@poczta.onet.pl> | 2023-06-10 18:13:50 +0200 |
---|---|---|
committer | Bartosz Kosiorek <gang65@poczta.onet.pl> | 2023-06-13 12:59:33 +0200 |
commit | cbb215aa20783523555185c83875ea5d5b94535b (patch) | |
tree | 68d31593174fe0c52557312d0fa9e0b5827a6890 /winaccessibility/source/service/AccEventListener.cxx | |
parent | 3e3d6aeec92a7252d92243e2de34b424cf5f7cdf (diff) |
tdf#143877 Fix failing tests caused by floating point precision
Due to different imlementation of floating-point unit (FPU),
on different CPU platforms, the floating point numbers could
could be different.
https://stackoverflow.com/questions/64036879/differing-floating-point-calculation-results-between-x86-64-and-armv8-2-a
https://mcuoneclipse.com/2019/03/29/be-aware-floating-point-operations-on-arm-cortex-m4f/
With this path I have changed the tested images,
to use floating point numbers which are easily represented
by floating numbers (multiplied/divided by 2), like:
- change tension to values: 0.125, 0.25, 0.5, 1.0, 1.5 ...
- change position of curve to of control points to 256.0, 384.0 512.0
Previous values was hard to represent by floating numbers,
for example tension:
- 0.4 has been written as 0.399999976158142
- 0.1 has been written as 0.099999994039535
More information:
https://observablehq.com/@benaubin/floating-point
Additionally the precision of numbers were
increased to double.
Change-Id: I5725c1f2f474d0c00821edaa9bb2102cb172093f
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/152838
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Tested-by: Bartosz Kosiorek <gang65@poczta.onet.pl>
Diffstat (limited to 'winaccessibility/source/service/AccEventListener.cxx')
0 files changed, 0 insertions, 0 deletions