Abstraction of cache and context #4357
Conversation
|
|
M1nd3r
force-pushed
the
M1nd3r/cache-and-context
branch
2 times, most recently
from
dbaba59
to
9400f19
Compare
[no changelog]
M1nd3r
force-pushed
the
M1nd3r/cache-and-context
branch
2 times, most recently
from
545adf2
to
48f58b0
Compare
M1nd3r
changed the title
M1nd3r
force-pushed
the
M1nd3r/cache-and-context
branch
2 times, most recently
from
addd607
to
3411ee1
Compare
[no changelog]
[no changelog]
M1nd3r
force-pushed
the
M1nd3r/cache-and-context
branch
from
3411ee1
to
f51e6dc
Compare
M1nd3r
changed the title
| @@ -557,6 +557,7 @@ if FROZEN: | |||
| )) | |||
|
|
|||
| SOURCE_PY.extend(Glob(SOURCE_PY_DIR + 'trezor/wire/*.py')) | |||
| SOURCE_PY.extend(Glob(SOURCE_PY_DIR + 'trezor/wire/codec/*.py')) | |||
There was a problem hiding this comment.
The codec specific files are separated so that they can be easily ommited for THP prod builds. When the THP implementation is added, it could look like this:
if THP:
SOURCE_PY.extend(Glob(SOURCE_PY_DIR + 'trezor/wire/thp/*.py'))
if not THP or PYOPT == '0':
SOURCE_PY.extend(Glob(SOURCE_PY_DIR + 'trezor/wire/codec/*.py'))
There was a problem hiding this comment.
Note: The condition or PYOPT == '0' is in the snippet because the codec will be used on DebugLink even with THP builds.
M1nd3r
requested review from
matejcik,
andrewkozlik,
onvej-sl and
prusnak
as code owners
…wnClass [no changelog]
core/src/storage/cache.py
Outdated
| @@ -156,210 +12,33 @@ def __init__(self) -> None: | |||
| # bytearrays, then later call `clear()` on all the existing objects, which resets them | |||
| # to zero length. This is producing some trash - `b[:]` allocates a slice. | |||
There was a problem hiding this comment.
The comment should go where the code below it went.
There was a problem hiding this comment.
Well, I thought about it previously. The problem with this "comment" is that it explains the process of allocation and the logic behind it.
- Allocation is two-step: first allocate, then clear.
- Note that using b[:] allocates a slice.
The allocation process actually happens just below where the commit is:
_SESSIONLESS_CACHE = ... # step 1
_PROTOCOL_CACHE.initialize() # step 1 and step 2
_SESSIONLESS_CACHE.clear() # step 2
The problem I see is that it is more abstract now - the "real" allocation (and use of slices) happens partly inside a _PROTOCOL_CACHE and SessionlessCache - that are in different files now.
I see a few solutions:
- keep it as it is now - it works as a general documentation
- remove the comment, move the implementation information into docs
- have the comment in each of the implementations (cache_codec, cache_thp)
- some mix of 2. and 3., divide the allocation notes and use relevant parts where necessary
@mmilata WDYT?
There was a problem hiding this comment.
I'd go with 3., some duplication won't hurt. Feel free to extend the docs if it makes sense to you.
| Filter = Callable[[int, Handler], Handler] | ||
|
|
||
| # If set to False protobuf messages marked with "experimental_message" option are rejected. | ||
| EXPERIMENTAL_ENABLED = False |
There was a problem hiding this comment.
Note to self: shouldn't this also be somewhere under storage/cache to avoid being wiped by a restart?
[no changelog]
| @@ -58,9 +58,6 @@ async def read( | |||
|
|
|||
| async def write(self, msg: protobuf.MessageType) -> None: ... | |||
|
|
|||
| def write_force(self, msg: protobuf.MessageType) -> Awaitable[None]: | |||
There was a problem hiding this comment.
Note: Removed write_force as it is not used by any code in main or in the PR. @mmilata
Abstraction of cache and context so that the current codec_v1 implementation can be switched for THP version.