Run custom tool before each build even if it is up-to-date - visual-studio

I want to embed the output of git describe --dirty --always --tags into my program to easily differentiate different versions of the same executable. To do so I generate a source file containing the output of said command.
Since the command produces a string that ends with dirty if there are any modified files in my worktree that are not yet committed I cannot really define a trigger when to re-generate my source file (e.g. as show in this answer).
To solve this problem I wrote a batch script which compares the content of the existing source file and the output of the git command and re-generates the source file only if necessary.
My problem now is that I need to run this batch script before each build, even if msbuild considers the project up-to-date. I tried to use a Pre-Build Event but it is not triggered if the project is up-to-date. I also tried a Custom Build Step and configured a non-existing file as output (this is also an accepted answer here). This causes my batch script to always execute but it also causes code generation being executed for the project even if my generated source file did not change.
4>------ Build started: Project: TestProject, Configuration: Release Win32 ------
4> Current git Version is 9.6-13-g2c753a1727-dirty
4> Include with version number is up-to-date
4> Generating code
4> 0 of 9106 functions ( 0.0%) were compiled, the rest were copied from previous compilation.
4> 0 functions were new in current compilation
4> 0 functions had inline decision re-evaluated but remain unchanged
4> Finished generating code
4> TestProject.vcxproj -> C:\bin\TestProject.exe
4> TestProject.vcxproj -> C:\bin\TestProject.pdb (Full PDB)
I tried it with Execute Before set to ClCompile. Might this be the cause? I also tried some others but with the same effect.
How can I always run my batch script but make the project only produce a new executable if my generated source file was updated?

How can I always run my batch script but make the project only produce
a new executable if my generated source file was updated?
Try adding <DisableFastUpToDateCheck>true</DisableFastUpToDateCheck> into PropertyGroup:
<PropertyGroup Label="Globals">
...
<DisableFastUpToDateCheck>true</DisableFastUpToDateCheck>
</PropertyGroup>
Then VS will always run the project, and if the output xx.exe is up-to-date with no changes to xx.cpp files, the ClCmpile target won't actually create new xx.exe.
Update:
I create a new simple C++ project, set it release mode, and add the DisableFastUpToDateCheck property. Now the vs will always build the project.
I both use the pre-build event and custom target which execute before ClCompile to execute the command git describe --dirty --always --tags.
Now my msbuild output log verbosity is minimal.
1.Build the project:
2.Change nothing to source file,build the project again:
3.Do the same action like #2. And same result.
4.Add a comment the xx.cpp(Not affect the code), build the project:
5.Msbuild project build output verbosity is a good tool when you try to get details about what happen in the build process. For me, I always set it to Detailed. For this issue, if we set it to Normal we can find:
In my opinion:
When I change something in code, the message is 1 of 10 functions (10.0%) were compiled, the rest were copied from previous compilation. And after comparing #1 and #4, we can conclude when it shows 0 of xxx functions were compiled, it actually not spend time generating code, so it won't affect time of your build process.
For the reason why it creates new executable, please see this document. In vs, most of the targets have their inputs and outputs. Like ClCompile target, xx.h files and xx.cpp files are its inputs. Whether this target will execute doesn't depend on if the output files are earlier than inputs.
For example: When I add a comment in xx.cpp, code actually not change. But since now the xx.cpp is newer than output xx.exe from earlier build. Then this time the build engine will execute ClCompile and Link target to create new xx.exe. And code generation is part of Link target, since no changes to code itself, it will show 0 of xxx functions were compiled to indicate it won't really generate though this time new xx.exe will be created.
So in normal situation the DisableFastUpToDateCheck do work for this situation like #2 and #3. According to the pic you shared above, you must do something which not change the code content(0 functions were compiled) in source files but update the modified date of input files.

Related

Xcode 10, sourcery & swiftlint build phases order

After updating to Xcode 10 there are some issues with initial project configurations. The steps look like this:
Generating some files using Sourcery
Linting with SwiftLint
Build
And configuration works like this:
And this was working fine in Xcode 9, but apparently, it's not working under Xcode 10 build system. The issue is if I download repository (.generated files are not a part of the repository). And hit build it will show me results like:
...
Using configuration file at '.sourcery.yml'
Scanning sources...
Found 239 types.
Loading templates...
Loaded 9 templates.
Generating code...
Finished.
Processing time 0.491948962211609 seconds
...
So sourcery works ok, then linting:
Linting 'FromResponse.generated.swift' (1/186)
Works fine as well as a process, but in the end, build will fail with:
error: Build input files cannot be found:
'/path/Generated/FromResponse.generated.swift',
...
So it's strange since a files are physically there and according to build settings file should be compiled later than generating them. How should I approach sourcery in build process then?
We are able to "fix" this in a similar way as this Github issue:
https://github.com/mac-cain13/R.swift/issues/438#issuecomment-416982928
We have to add each of the generated files by Sourcery to the output folder and it will get picked up by the Xcode correctly.
This relates to changes in the New Xcode Build System that has been enabled by default in Xcode 10. If a build phase creates files which are needed as an input to a later build phase then it needs to specify them explicitly in the Output Files or Output File List.
In your example the Sourcery build phase is generating Swift source file(s) which are needed as input to the Compile Sources phase.
However, the issue at this point is that after you specify the output files for Sourcery, the build phase doesn't re-run every time as it sees the output file is already there. So far I haven't worked out a useable solution to this part, beyond running tools like Sourcery and SwiftGen manually or keeping the generated files in Git so that they are always present.
I was having a similar problem with a buildphase calling mogenerator to build my data model classes. The approach given by #Yuchen should work.
#Andrew: To force always (re-)running this build phase, I add the line
touch .alwaysRun
as the last line in the shell command to run and mark the build phase "Input Files" to have $(SRCROOT)/.alwaysRun. That seems to do the trick.

How to build to same folder each time

In TFS the Build number format usually looks something like this:
$(BuildDefinitionName)_$(Date:yyyyMMdd)$(Rev:.r)
However, I only want to retain 1 build and I would like it to build to the same folder each time. So I tried changing it to this:
$(BuildDefinitionName)
But the problem is that this only works one time, then gives an error that the build number already exists after that. I would like to build to the same folder so that I can write a script to zip the latest build, move it to another place, and then unzip it and it would just be much easier if I didn't have to deal with writing code to figure out what the most recent folder name is.
Is there a way to accomplish building to a folder name that doesn't change?
This is by designed, every completed build should has a unique build number/name. Otherwise you will get the error above.
$(Rev:.r)
Use $(Rev:.rr) to ensure that every completed build has a unique
name. When a build is completed, if nothing else in the build number
has changed, the Rev integer value is incremented by one.
As a workaround: For vNext build, you could use a copy Files task to copy the build output to the same folder during the build pipeline. To make sure you will always only get the latest build, you could add a powershell script before the copy task to clean/delete files in that special target folder.
For XAML build you need to customize the build template and add the default CopyDirectory activity in build template to copy the build result to the specified drop location. The detailed steps please refer to this blog. Also add a pre-build script to do the clean operation.

How can I dynamically set my app's build number without dirtying my source tree?

I'm using git-svn and I'm trying to embed my revision number into my iOS app. At the moment, I have a build phase which runs the following script:
SVN_REVISION=$(git svn find-rev HEAD)
/usr/libexec/PlistBuddy -c "Set :CFBundleVersion $SVN_REVISION" "$INFOPLIST_FILE"
The problem with this is that, since the repo always contains the previous revision, the script always makes my Info.plist dirty.
Is it possible for me to dynamically set my app's build number without dirtying my source tree?
1) Add a new target to your project of type "Aggregate", e.g. you may name it "Update Info.plist Prefix Header"; just use that as "Product Name" in the dialog.
2) Add a Run Script build phase to this new target with the following source code:
#!/bin/sh
SVN_REVISION=$(git svn find-rev HEAD)
echo "#define SVN_REVISION $SVN_REVISION" > "$SCRIPT_OUTPUT_FILE_0"
3) Add an output file to your script, name it
$(CONFIGURATION_TEMP_DIR)/InfoPlist.pch
4) Open the Build Phases of your iOS app.
5) Add the aggregate target you created before as dependent target (add it to "Target Dependencies"). This means Xcode will always first build this target before it will build your iOS target.
6) Open the Build Settings of your iOS app.
7) Search for the setting "Info.plist Preprocessor Prefix File" and change it to exactly the same value you used for the output file in step (3).
8) Search for the setting "Preprocess Info.plist File" and make sure it is enabled.
9) Open your current Info.plist file and change the value of CFBundleVersion to SVN_REVISION. Do not use $(SVN_REVISION) or ${SVN_REVISION}; this is no build setting or environment variable replacement, this is a preprocessor replacement, so just use SVN_REVISION.
That's it. Each time you build your iOS app, Xcode first builds the aggregate target, which updates the PCH file, and when it builds your iOS app, it will run the Info.plist file through the C pre-processor (using the PCH file as prefix header) before copying it to your application. The pre-processor will replace SVN_REVISION since it is defined as a macro in your PCH file.
Important Notes
Some people may think it is a better idea to use $(DERIVED_FILE_DIR) instead of $(CONFIGURATION_TEMP_DIR). Well, in theory they are right, yet there is just one problem in practice: The derived file dir is different for every target, while the configuration temp dir is the same (it is only different for every build configuration). When using derived file dir, the PCH file is written to the derived file dir of the aggregate target, yet when building the iOS app, Xcode will search for this file in the derived file dir of the iOS app and thus it won't find the file.
Some people may also think it is a better idea to just add the Run Script phase that updates the prefix header as the first build phase of you iOS app instead of creating a separated target for it (this would also resolve the derived file dir issue mentioned above). Again, nice idea in theory but cannot work in practice: If preprocessing is requested, the Info.plist is preprocessed before the first script phase is even executed, so if the PCH file does not exist already or has not been updated already, either the build terminates with an error or an outdated SVN revision is written to the plist file. That's why you need a separate target for this task that is guaranteed to be build before your actual target is.
Mecki thank you for the excellent answer! I applied the same concept to set a version timestamp and the current git SHA for the build.
FYI I just ran into a small issue. It seems that, at least in Xcode 5, if you specify an output file the script step uses it as a cache, so no matter the changes I made to my actual app code the script reported that it had ran but the values were not the current ones...
I had to move the output file declaration to the script itself to solve the issue, i.e. added
SCRIPT_OUTPUT_FILE_0="$CONFIGURATION_TEMP_DIR/InfoPlist.pch"
to the top of my script.
Additionally the original plist should also be touched in order for the build step to copy the new values in, so I also added
`touch $SCRIPT_INPUT_FILE_0`
after the previous output file declaration. This touch operation does not make git detect the change as commit-able.
Cheers

Which MSBuild tasks initiate a call to SGen?

I have inherited a complex MSBuild script which builds & deploys multiple solutions. When I run the script I see the following error in log:
SGEN: Cannot generate serialization assembly
C:\B\268\Agents\Agents.XmlSerializers.dll
because it already exists. Use /force to force an overwrite of the
existing assembly.
I agree that Agents.XmlSerializers.dll exists. What I can't determine is which task caused it to be created and which one initiated the attempt to re-create it.
I have searched through the files in the build hierarchy for 'SGen' and found nothing. So I presume that SGen is called implicitly by two (or more) of the tasks in the build.
Can anyone point out what might call SGen?
Thanks.
In the Build Output window, look for...
Task "SGen"
Right above that is the actual "sgen.exe" command.
And as Pawell said, you need full details in your build output. Inside of Studio, it's Tools - Options... - Projects and Solutions - Build and Run - MSBuild project build output verbosity: Diagnostic.

Xcode: Running a script before every build that modifies source code directly

What I did:
I have a script that
Read some configuration files to generate source code snippets
Find relevant Objective-C source files and
Replace some portions of the source code with the generated code in step 1.
and a Makefile that has a special timestamp file as a make target and the configuration files as target sources:
SRC = $(shell find ../config -iname "*.txt")
STAMP = $(PROJECT_TEMP_DIR)/$(CONFIGURATION)$(EFFECTIVE_PLATFORM_NAME).stamp
$(STAMP): $(SRC)
python inject.py
touch $(STAMP)
I added this Makefile as a "Run Script Build Phase" on top of the stack of build phases for the project target.
What happened:
The script build phase was run before compiling the source.
However, since the script modifies source code during its execution, I needed to build twice to get the most recent version of the build product. Here is what I imagine to be happening:
1st run: Xcode collects dependency information ---> no changes
1st run: Xcode runs "Run Script Build Phase" ---> source is changed behind Xcode's back
1st run: Xcode finishes build, thinking nothing needs to be updated
2nd run: Xcode collects dependency information ---> source has changed, needs rebuild!
2nd run: Xcode runs Run Script Build Phase" ---> everything is up-to-date
2nd run: Xcode proceeds to compilation
After reading Xcode documentation on Build Phases, I tried adding a source file which is known to be updated every time the script is run as the output of "Run Script Build Phases", but nothing changed. Since the number of configuration files may vary in my project, I don't want to specify every input and output file.
Question:
How do I make Xcode aware of source file changes made during "Run Script Build Phase"?
Edit:
Added that I placed the script build phase before the other build phases
Every technique mentioned so far is an overkill. Reproducing steve kim's comment for visibility:
In the build phases tab, simply drag the "Run Script" step to a higher location (e.g. before "Compile Sources").
Tested on Xcode 6
This solution is probably outdated. See the higher voted answer instead; I no longer actively use Xcode and am not qualified to vet a solution.
Use "External Target":
Select "Project" > "New Target..." from the menu
Select "Mac OS X" > "Other" > "External Target" and add it to your project
Open its settings and fill in your script setup
Open the "General" tab of the main target's settings and add the new target as it's direct dependency
Now the new "External Target" runs before the main target even starts gathering dependency information, so that any changes made during the script execution should be included in the build.
There is another, slightly simpler option that doesn't require a separate target, but it's only viable if your script tends to modify the same source files every time.
First, here's a brief explanation for anyone who's confused about why Xcode sometimes requires you to build twice (or do a clean build) to see certain changes reflected in your target app. Xcode compiles a source file if the object file it produces is missing, or if the object file's last-modified date is earlier than the source file's last-modified date was at the beginning of the first build phase. If your project runs a script that modifies a source file in a pre-compilation build phase, Xcode won't notice that the source file's last-modified date has changed, so it won't bother to recompile it. It's only when you build the project a second time that Xcode will notice the date change and recompile the file.
Here's a simple solution if your script happens to modify the same source files every time. Just add a Run Script build phase at the end of your build process like this:
touch Classes/FirstModifiedFile.m Classes/SecondModifiedFile.m
exit $?
Running touch on these source files at the end of your build process guarantees that they will always have a later last-modified date than their object files, so Xcode will recompile them every time.
As of Xcode 4, it looks like if you add the generated files to the output section of the build phase, it will respect that setting, and not generate the ... has been modified since the precompiled header was built error messages.
This is a good option if your script is only generating a handful of files each time.
I as well struggled with this for a long time. The answer is to use ento's "External Target" solution. He is WHY this problem occurs and how we use it in practice...
Xcode 4 build steps do not execute until AFTER the plist has been compiled. This is silly, of course, because it means that any pre-build steps that modify the plist won't take effect. But if you think about it, they actually DO take effect...on the NEXT build. That's why some people have talked about "caching" of plist values or "I have to do 2 builds to make it work." What happens is the plist is built, then your script runs. Next time you build, the plist builds using your modified files, hence the second build.
ento's solution is the one way I've found to actually do a true pre-build step. Unfortunately I also found that it does not cause the plist to update without a clean build and I fixed that. Here is how we have data-driven user values in the plist:
Add an External Build System project that points to a python script and passes some arguments
Add user-defined build settings to the build. These are the arguments that you pass to python (more on why we do this later)
The python script reads some input JSON files and builds a plist preprocessor header file AND touches the main app plist
The main project has "preprocess plist files" turned on and points to this preprocessor file
Using touch on the main app plist file causes the main target to generate the plist every time. The reason we pass in build settings as parameters is so our command-line build can override settings:
Add a user-defined variable "foo" to the prebuild project.
In your prebuild you can use $(foo) to pass the value into the python script.
On the command-line you can add foo=test to pass in a new value.
The python script uses base settings files and allows for user-defined settings files to override the defaults. You make a change and immediately it ends up in the plist. We only use this for settings that MUST be in the plist. For anything else it's a waste of effort....generate a json file or something similar instead and load it at run-time :)
I hope this helps...it's been a couple rough days figuring this out.
The External Target solution from #ento no longer works as of Xcode 11.5. The solution is to add all files that will be changed under Output Files in the Run Script.
Another option is to create a subproject framework with your scripts and just add it as a dependency to all targets. The phase scripts of this subproject should now be executed before all targets.

Resources