Fixed
Status Update
No update yet.
Comments
je...@google.com
what is in update.zip ? can you share that file to triage this issue
sh...@google.com
MHC19Q for Nexus 5X (bullhead) - https://dl.google.com/dl/android/aosp//bullhead-ota-mhc19q-8fe67a2b.zip
N4F26T for Nexus 5X (bullhead) -https://dl.google.com/dl/android/aosp/bullhead-ota-n4f26t-648ce802.zip
WW-12.2.5.23 for Asus ZenFone Go (ASUS_X014D) -http://dlcdnet.asus.com/pub/ASUS/ZenFone/ZB452KG/UL-ASUS_X014D-WW-12.2.5.23-user.zip
Look like, this happens only with a big images.
N4F26T for Nexus 5X (bullhead) -
WW-12.2.5.23 for Asus ZenFone Go (ASUS_X014D) -
Look like, this happens only with a big images.
je...@google.com
ap...@google.com
related to (and probably worked around for now by) internal bug http://b/36046324
i do think we should also start shipping a 64-bit Windows platform tools package.
we should also consider changing android::base::ReadFdToString to call fstat and pre-size the vector. not worthwhile for reading things like /proc/uptime, but the default expansion behavior for a huge file like a full OTA update is going to result in substantial overhead. (which will explain the regression between .3 and .4, since the former will have allocated exactly the necessary number of bytes.)
lastly (and maybe not worth doing at all, depending on how soon we can get folks to use a 64-bit platform tools on Windows, since all the other platforms are already 64-bit-only), we could consider rewriting the adb sideload code to (a) mmap/munmap rather than actually read in to physical memory or (b) pread. i'm not sure what the performance impact of pread would be (especially on Windows where there is no pread equivalent).
i do think we should also start shipping a 64-bit Windows platform tools package.
we should also consider changing android::base::ReadFdToString to call fstat and pre-size the vector. not worthwhile for reading things like /proc/uptime, but the default expansion behavior for a huge file like a full OTA update is going to result in substantial overhead. (which will explain the regression between .3 and .4, since the former will have allocated exactly the necessary number of bytes.)
lastly (and maybe not worth doing at all, depending on how soon we can get folks to use a 64-bit platform tools on Windows, since all the other platforms are already 64-bit-only), we could consider rewriting the adb sideload code to (a) mmap/munmap rather than actually read in to physical memory or (b) pread. i'm not sure what the performance impact of pread would be (especially on Windows where there is no pread equivalent).
je...@google.com
Correction to #5: Windows does support pread() - it's in ReadFile()'s OVERLAPPED parameter.
ap...@google.com
@7: we'd still need to write the pread wrapper, though. (similar if we went with mmap/munmap.)
https://android-review.googlesource.com/#/c/355481/ pre-sizes the vector, which should be enough to undo the regression from '.3 to '.4...
Description
The reasoning is that the padding defines the content rectangle, and whether or not children's pixels are clipped to that rectangle or not doesn't change the semantics of that rectangle.
Consider for example a horizontal ViewPager2 that has unequal padding on the left and the right. When it is in idle state, a page exactly fills the rectangle described by the inner edge of the padding, so the page looks to be off-center by design. When snapping kicks in (e.g., because of a swipe), the page to which it is snapping should end up in that same "off-center" position, exactly matching the rectangle described by the inner edge of the padding, regardless of the value of clipToPadding.