For aosp_marlin (on AOSP), this speeds up target files creation from
~57 seconds to ~15 seconds.
For marlin-userdebug (on internal master), this speeds up target files
creation from ~300 seconds to ~95 seconds.
This does make some minor changes to the resulting target_files zip:
* All of the META files are actually at the beginning of the zip now,
previously the ones created by add_img_to_target_files would be at
the end.
* The images are slightly larger. Go's zip implementation in soong_zip
isn't quite as good as the `zip` tool.
* vendor_filesystem_config.txt isn't present if we're not building a
vendor image.
Test: aosp_marlin target_files zip changes look reasonable
Test: marlin target_files zip changes look reasonable
Test: bullhead target_files zip changes look reasonable
Change-Id: Ia5ce6870b85559d65107c3b94332391b4b43a0ea
In addition to the current behavior of add_img_to_target_files working
on an existing zip file, allow passing in a directory where the target
files have already been extracted. When in this mode, it writes the
images out to that directory instead of the zip file.
This allows us to call add_img_to_target_files on the temp directory
used during the build to create the target files package, saving the
time and space of unzipping what we just zipped. This also allows us to
use the parallel soong_zip, which compresses the images much faster.
Test: aosp_marlin target_files zip is the same before/after this change
Test: marlin target_files zip is the same before/after this change
Test: bullhead target_files zip is the same before/after this change
Change-Id: I155654cdc7ad7754ba4ef76ec69c31f504a58198
We have stopped shipping devices with file-based OTAs, and are not
actively maintaining the support. Devices using file-based OTAs
should be moved to block-based, if not A/B OTAs.
We will also need to clean up EdifyGenerator class, which will be
handled in follow-up CLs.
Bug: 35853185
Test: Generate full and incremental OTAs w/ and w/o the CL, and they
give identical packages.
Test: Not specifying --block also generates block-base OTAs.
Change-Id: I3b0fc8ce5600e109f3251fe41f655534aaa298c7
We must have created the images (system.img, system.map etc) prior to
calling ota_from_target_files.py (added by commit
2e0d8fcf08, into Lollipop).
Remove the obsolete suppport for handling "old" target_files zips that
don't have such images. This simplies the logic for BuildSystem() and
BuildVendor(), which now would only be called by
add_img_to_target_files.py itself.
Test: Generating full and incremental OTAs give the same results, w/ and
w/o this CL.
Change-Id: I0ea243d09d3378249d7982701ae4ec339b04b7b8
Otherwise it doesn't work for incremental builds.
Bug: 34929159
Bug: 35997300
Test: repo sync && m
Change-Id: I78ead57cf89a9d0f26dbf4f3a225a9357bdca1b3
The first one in WriteVerifyPackage() is mismatching function parameters
that can be trivially fixed.
The other one is in WriteABOTAPackageWithBrilloScript(), where we don't
have edify script instance.
Test: `pylint --rcfile=pylintrc ota_from_target_files.py`.
Change-Id: Ie238ef5b296dfe9e725b61376992447b662d2376
The major issue with the existing implementation is unnecessarily
holding too much data in memory, such as HashBlocks() which first reads
in *all* the data to a list before hashing. We can leverage generator
functions to stream such operations.
This CL makes the following changes to reduce the peak memory use.
- Adding RangeSha1() and WriteRangeDataToFd() to Image classes. These
functions perform the operations on-the-fly.
- Caching the computed SHA-1 values for a Transfer instance.
As a result, this CL reduces the peak memory use by ~80% (e.g. reducing
from 5.85GB to 1.16GB for the same incremental, as shown by "Maximum
resident set size" from `/usr/bin/time -v`). It also effectively
improves the (package generation) performance by ~30%.
Bug: 35768998
Bug: 32312123
Test: Generating the same incremental w/ and w/o the CL give identical
output packages.
Change-Id: Ia5c6314b41da73dd6fe1dbe2ca81bbd89b517cec
This reverts commit a7316ce094.
This CL differs from the original CL by not unzipping RADIO/*. This is
because: a) AOSP targets don't have RADIO/ entries in the TF.zip; b)
we're not using the unzipped RADIO files (but reading them from the zip
files directly) - checked all the device-specific releasetools for
angler, bullhead, ryu, shamu, volantis, fugu, marlin and sailfish.
Test: `m dist` with AOSP targets (tested fugu and bullhead).
Change-Id: I4d0c67214ddd6202fc27c71bb79f52b5f4d40c64
This reverts commit aa3a04f19d.
Reason for revert: Some AOSP targets don't include RADIO/ in the TF.zip. We may possibly skip unzipping RADIO/, or by always creating a dummy RADIO folder in TF.zip. Revert this CL for now.
Change-Id: I8e90d322706a4fe82129bdfab5ffa1eab989c648
Support BOARD_VENDOR_KERNEL_MODULES which accepts a list of the
kernel module binaries. Binaries listed there are copied to
$(TARGET_OUT_VENDOR)/lib/modules and 'depmod' is executed to generate
the dependencies. BOARD_RECOVERY_KERNEL_MODULES is similarly
implemented to provide a way to specify kernel modules to be copied
to the recovery image.
Bug: 33305644
Test: android master build on marlin and load dummy modules
Merged-In: I0a58a270dbb0d2d38fd089f4abff66e62e14fac1
Change-Id: Ia5f429508ca3d7c77448358e42af6efa98fa8624
When building BBOTAs, it only needs *some* unzipped entries in the given
target_files zip(s). In particular, it needs 'IMAGES/*', 'META/*',
'RADIO/*'. (It also reads 'SYSTEM/build.prop' and 'OTA/bin/updater', but
directly from the zip file.)
This CL specifies the entries to unzip. It saves the I/O cost, as well as
the temporary storage.
Test: ota_from_target_files.py gives the same package w/ and w/o the CL.
Test: check_target_files_signatures.py still works.
Change-Id: I728428aa0e138879e49f9efbdb46a85892fc7038
It is not necessary to have it in setup_one_odex since the profile is
the same for multilib.
Prevents errors related to using ../ in profile path resulting in not
having a directory.
Bug: 34929159
Test: m
Change-Id: Ifdcba119f3599b0c80e27594781db5e4d64bd473
Tao Bao