IssueTracker

This doesn't repro for me. Steps:
(1) Create new C++ project
(2) Run the app (this create's the binary directories
(3) Do something to cause a gradle Sync
(4) Look for files under .externalNativeBuild. They're there:
    jomof@jomof:~/projects/client/src/android/MyApplication4/app$ find "." -name "*.o"
    ./.externalNativeBuild/cmake/debug/x86/CMakeFiles/native-lib.dir/src/main/cpp/native-lib.cpp.o

Please add repro steps so I can get a better idea of what's happening.

>>  It should only delete the CMake binary directories *if and only if* parameters change in Gradle that cause cmake toolchain parameters to change (such as platform version, abi, toolchain) or parameters passed to CMake to change.
This is the way it's design to work and it is working that way for the cases I know

I realized that there were a few things we were doing that were questionable. Specifically, we set up our CMake command line params in a variable at the root build.gradle script. We were generating build dates and such in gradle and passing those down to CMake. I think things like this may be causing the CMake args to change too often, possibly resulting in this behavior. I'm in the process of eliminating a lot of this, so I'll make sure to report back the results.

So even after making my changes, I still see this behavior. My repro steps:

1. Make a change to a CMake script
2. Switch to Android Studio. I am prompted to do gradle sync, and so I do it.

On next build, a full rebuild is performed. For some reason it is not performing a delta build. Some libraries are rebuilding that should not be rebuilt. So obviously a "ninja clean" is happening somewhere. There's simply no other explanation for this.

Jomo, can you please reopen? This is still a very real problem.

Hey Robert,
From what you wrote I think this is intended behavior. If you change the CMake script then the CMake-generated Ninja project must be regenerated. Since there's a new Ninja project there needs to be a full build. There's no concept of delta build that survives changes to build command files (CMakeLists.txt or ninja build files) because any behavior difference is possible.

Most of the time, a CMakeLists.txt is modified to add, remove, or rename source files. CMake uses this list of source files as a way to know if it should regenerate to update its ninja scripts to build the new files, if new files were added.

If you do a traditional CMake build from the command line, CMake will regenerate but it will not clean your previously compiled translation units (the object files). Those remain in-place. This allows you to perform another build (e.g. execute "ninja" command) and the build tool and compiler working together will be smart enough to know which files need to be rebuilt.

From this standpoint, I disagree with your assertions. There's no functional reason AFAIC to "clean" build every time a CMake script changes. If command line parameters to CMake change in such a way that it would completely invalidate previously compiled translation units outside of the knowledge of ninja and the toolchain programs, such as changing the Android ABI via toolchain arguments, then yes you would need to clean everything and rebuild. However, beyond that, you should be relying on ninja and the compiler to perform delta builds on previously compiled translation units.

As you have it now, I am unable to do a simple task like add new CPP files without having to rebuild my entire project. This wastes about 15-20 minutes of my time (that's how long a full rebuild takes), versus less than 60 seconds to just build the new file.

I think you should aim for smarter introspection into how this mechanism works. I see two very simple scenarios:

1) A build & sync was triggered due to parameter changes from gradle (those that affect the externalNativeBuild configuration, such as native ABI changes and custom user arguments)
2) A build was triggered because CMake.exe told you that a regeneration was needed since it detected a CMakeLists.txt changed

#1 should perform a clean & rebuild
#2 should not require any special action from you; you should simply run "ninja" command as if the scripts hadn't changed at all, and let Ninja run CMake to regenerate as needed

I feel like #2 is being treated as #1 right now. Also I noticed that Android Studio wants to re-sync Gradle when a CMakeLists.txt file changed. I do not think Android Studio should be "watching" CMake scripts. It should only resync if gradle scripts changed. If a CMake script changes, let your "ninja" command handle it.

As a final note, I realize I'm probably oversimplifying this a little. I'm sure there's reasons and edge cases as to why you need that level of control over CMake. However, at some point the full clean build has to be triggered in a smarter way to avoid these giant productivity problems. You obviously know much more about how CMake integrates with Android Studio than I do; I can only offer feedback based on my direct usage of CMake outside of AS. However, I hope my feedback makes sense and is agreeable.

Just to clarify, in the logs this is the exact problem:

External native generate JSON x86Debug: - dependent build file missing or changed
External native generate JSON x86Debug: removing stale contents from 'E:\code\frontend\source\Applications\zPayService\TestTool\.externalNativeBuild\cmake\x86Debug\x86'

The "removing stale contents"... those contents aren't stale at all. A build could be executed that reuses those existing object files if all I did was modify the cmake script to add new files to compile.

Also CMake will think CMake scripts changed if you swap between git branches:

1. Go from branch A to branch B
2. Go from branch B back to A

Cmake script is still physically the same, but because you swapped branches around, the timestamp of the Cmake script changed so this triggers a full rebuild.

At this point I'd say this is our biggest issue at Ziosk and the one causing the most productivity problems. Our team of about 15 engineers is constantly complaining. There's concern on how long Google will take to address these issues, some have proposed rolling back to Eclipse + ANT. If we're a year out from any improvements I may have to consider it.

I'm prototyping a change to address this. There is a concern that CMake doesn't operate incrementally under certain circumstances. For example, changing the compiler. I think Refresh Linked C++ Projects should be enough to cover this case. I'm still looking for other cases that would be more concerning

Another kind of change to CMakeLists.txt is affected. If a build target is removed then CMake won't delete the .so and related .o files. Gradle packages *.so rather than the explicit list of .so files because build systems sometimes copy extra .so files into the output folder to be packaged.
I think I can cover this case by specific removing only unrecognized .so files during sync.

To your example: Changing compiler means changing command line arguments (e.g. `-G Ninja`) or alternatively, changing environment such as C and CXX variables. In the latter case, normal CMake does not watch for changes to environment variables, so it's the responsibility of the user (or in your case, the IDE) to monitor changes to environment and control CMake accordingly.

However, this still falls into the pattern of behavior I've outlined previously. Indirect or direct changes to Gradle or Gradle scripts that result in a different command line invocation of CMake should result in a full clean (delete binary dir + regenerate using CMake). Changes to the CMake scripts themselves shouldn't be monitored by Android Studio at all. If you have weird cases where CMake is doing work the NDK toolchain should be doing, such as lots of system introspection and runtime changes to toolchain, I'd simply document that Google does not support that usage of CMake, and all changes should go through Gradle instead (or file a bug with the NDK team and have them fix the toolchain file).

Concerning stale *.so files: What I did for my ANT integration custom targets in CMake was to delete all *.so files before running a build, that way when a build is finished, you're left with only *.so files that were copied to that destination by custom commands tied to real targets. So if a target was completely removed, its corresponding *.so file would not make it to the temporary/intermediate copy destination where you look for *.so files to package into the APK.

This solution is simple and doesn't require introspection into CMake targets. If you have more direct access to targets defined by CMake, you could keep track a list of targets found during configuration and query the OUTPUT property on each of them and correlate that to files you find in your CMake output directory. Not sure if this has much benefit over the idea above, though.

I've submitted a fix for this. Unless we find a problem the fix will appear in the next 3.1 canary

Could you explain the fix here so I (and others) can get a good understanding of how the behavior is changing as well as what the root cause (original behavior) was?

All these years later and it's still not a there?

I need this!

It's been almost 4 years since this issue was raised. What's holding up such a small but important feature? Eric?

I agree ! I need this feature too !

I am on the same page - I urgently need this feature too!

Google Docs is a great program but it is very discouraging how little attention has been paid to updating it.  Perhaps a paid version?  I'd gladly pitch in if it meant improved functionality.

Replying to this email means your email address will be shared with the
team that works on this product.
Does that mean that the team is actually working on it?
Great!



On Mon, Apr 3, 2017 at 2:55 AM, <buganizer-system@google.com> wrote:

Are there any updates on this feature?

Any updates on this feature? It seems like getting the sheet either by the Name or the Index are poor methods because Names can be changed by users and Indexes can be changed by adding, removing or moving sheets. getSheetById() would give us a stable way of always accessing the exact same sheet.

Wow, it's almost 10 years...
I am not sure but it seems like a one-hour feature as we already have getSheetByName

They've likely decided for reasons not to implement this. It might go against the way they intend for apps script to function with documents. That would explain the methods that are available for getting sheets vs. the methods that aren't available.

Whatever the case, this is easy enough to work around with the solution proposed in the first post. I'd recommend any future users go that route; this is just one of a few different topics that have brought this up over the past decade. Unlikely change is coming soon.

Scrolled down to check the date of latest comment...wasn't disappointed.
Anyway for those looking for a less intuitive and not 100% solution: use openByUrl + getActiveSheet
A link that links directly to a certain sheet in a spreadsheet looks like this
https://docs.google.com/spreadsheets/d/[spreadsheet_id]/edit#gid=[sheet_id]
So if you open spreadsheet by this link, the active sheet would be exactly the sheet you are looking for

I am surprised there are only a few comments on this topic. I think that is another reason google hasn't created a method for us.

Wow only waiting 10 years for an explanation as to why this doesn't exist...

any news for this topic ?

wow, im still wondering why they haven't implemented this feature yet.

>  use openByUrl + getActiveSheet A link that links directly to a certain sheet in a spreadsheet looks like this https://docs.google.com/spreadsheets/d/[spreadsheet_id]/edit#gid=[sheet_id] 

You will be surprised when you find out that it doesn't work.



We needed this "getSheetById" method

Just tried right now and doesn't seem to be working yet.