[
https://issues.apache.org/jira/browse/MINIFICPP-2346?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
James Guzman (Medel) updated MINIFICPP-2346:
--------------------------------------------
Description:
*Description of Issue:* Building MiNiFi C++ with just CMake is really slow.
During the the build process, it clones all the external dependencies and then
builds those external dependencies while its building MiNiFi CPP. Sometimes the
MiNiFi CPP build fails too when external lib dependencies fail to download or
fail to build. It would be faster to build MiNiFi C++ with the external
dependencies already preinstalled.
*Potential Solution:* Integrate *conan version 2* into MiNiFi C++ project to
enable support for using *conan install* to install all or most of MiNiFi's
external dependencies, using *conan install --build=missing* to build them as
prebuilt binary conan packages, upload them to conancenter with {*}conan
upload{*}, and then run cmake generate to generate the appropriate build files
for the OS, then run make to build MiNiFi CPP. Also conan has really good
support for cross platform compatibility for Linux and Windows. At this point
because we already have most of MiNiFi CPP's external lib dependencies
installed with conan, our CMake build will just focus on building the MiNiFi
CPP code, enabling faster builds. Also I will need to account for backward
compatibility support with the previous way of installing MiNiFi CPP external
lib dependencies using CMake FetchContent_Declare(...) approach until
conanfile.py approach covers installing all these dependencies. A valid
verification of the MiNiFi C++ build would be to run GTESTs and if all the
expected test cases pass, then it may be fair to conclude we would be ready to
deprecate the slow CMake FetchContent_Declare(...) approach.
*Steps to Perform MiNiFi C++ Build Enhancement (Plan to Integrate into
script(s)):*
# Install conan version 2 into MiNiFi C++ dev environment (my preference is
using docker container)
# Create a conanfile.py file at root of MiNiFi C++ project where we run "conan
install" or "conan install --build=missing" to install MiNiFi C++ external lib
dependencies first. Then we run CMake to build MiNiFi C++.
# Find all MiNiFi C++ find_package(...) CMake function calls and check if
there is already supported conan packages on the conancenter, so we can install
those packages.
# Find all FetchContent_Declare(...) CMake function calls and check if there
is already supported conan packages on the conancenter, so we can install those
packages.
# At this point we will have used conan version 2 C++ package manager to
install almost all MiNiFi C++'s external dependencies.
## With Conan, we will have installed these MiNiFi external lib dependencies
as prebuilt binary conan packages.
## Or we will have instructed conan to first build each C++ external lib
dependency, upload each lib to conancenter.
# Thus, conan manages installing MiNiFi C++'s external lib dependencies, so we
can then run CMake to build MiNiFi faster.
Here is my draft PR for this Jira Ticket for your reference:
[https://github.com/apache/nifi-minifi-cpp/pull/1775]
* I have added *abseil* external lib dependency to conanfile.py and integrated
it into Abseil.cmake with conditional statement that checks if
USE_CONAN_PACKAGER minifi cpp option is true else checks if
USE_CMAKE_FETCH_CONTENT option is true for backward compatibility, so we can
began the process of integrating in conan version 2.
was:
*Description of Issue:* Building MiNiFi C++ with just CMake is really slow.
During the the build process, it clones all the external dependencies and then
builds those external dependencies while its building MiNiFi CPP. Sometimes the
MiNiFi CPP build fails too when external lib dependencies fail to download or
fail to build. It would be faster to build MiNiFi C++ with the external
dependencies already preinstalled.
*Potential Solution:* Integrate *conan version 2* into MiNiFi C++ project to
enable support for using *conan install* to install all or most of MiNiFi's
external dependencies, using *conan install --build=missing* to build them as
prebuilt binary conan packages, upload them to conancenter with {*}conan
upload{*}, and then run cmake generate to generate the appropriate build files
for the OS, then run make to build MiNiFi CPP. Also conan has really good
support for cross platform compatibility for Linux and Windows. At this point
because we already have most of MiNiFi CPP's external lib dependencies
installed with conan, our CMake build will just focus on building the MiNiFi
CPP code, enabling faster builds. Also I will need to account for backward
compatibility support with the previous way of installing MiNiFi CPP external
lib dependencies using CMake FetchContent_Declare(...) approach until
conanfile.py approach covers installing all these dependencies. A valid
verification of the MiNiFi C++ build would be to run GTESTs and if all the
expected test cases pass, then it may be fair to conclude we would be ready to
deprecate the slow CMake FetchContent_Declare(...) approach.
*Steps to Perform MiNiFi C++ Build Enhancement (Plan to Integrate into
script(s)):*
# Install conan version 2 into MiNiFi C++ dev environment (my preference is
using docker container)
# Create a conanfile.py file at root of MiNiFi C++ project where we run "conan
install" or "conan install --build=missing" to install MiNiFi C++ external lib
dependencies first. Then we run CMake to build MiNiFi C++.
# Find all MiNiFi C++ find_package(...) CMake function calls and check if
there is already supported conan packages on the conancenter, so we can install
those packages.
# Find all FetchContent_Declare(...) CMake function calls and check if there
is already supported conan packages on the conancenter, so we can install those
packages.
# At this point we will have used conan version 2 C++ package manager to
install almost all MiNiFi C++'s external dependencies.
## With Conan, we will have installed these MiNiFi external lib dependencies
as prebuilt binary conan packages.
## Or we will have instructed conan to first build each C++ external lib
dependency, upload each lib to conancenter.
# Thus, conan manages installing MiNiFi C++'s external lib dependencies, so we
can then run CMake to build MiNiFi faster.
> Speed up MiNiFi Build with Conan package manager (C++)
> ------------------------------------------------------
>
> Key: MINIFICPP-2346
> URL: https://issues.apache.org/jira/browse/MINIFICPP-2346
> Project: Apache NiFi MiNiFi C++
> Issue Type: Improvement
> Affects Versions: 0.15.0
> Environment: Ubuntu 22.04
> Reporter: James Guzman (Medel)
> Assignee: James Guzman (Medel)
> Priority: Minor
> Labels: build, conan, ubuntu
> Time Spent: 10m
> Remaining Estimate: 0h
>
> *Description of Issue:* Building MiNiFi C++ with just CMake is really slow.
> During the the build process, it clones all the external dependencies and
> then builds those external dependencies while its building MiNiFi CPP.
> Sometimes the MiNiFi CPP build fails too when external lib dependencies fail
> to download or fail to build. It would be faster to build MiNiFi C++ with the
> external dependencies already preinstalled.
>
> *Potential Solution:* Integrate *conan version 2* into MiNiFi C++ project to
> enable support for using *conan install* to install all or most of MiNiFi's
> external dependencies, using *conan install --build=missing* to build them as
> prebuilt binary conan packages, upload them to conancenter with {*}conan
> upload{*}, and then run cmake generate to generate the appropriate build
> files for the OS, then run make to build MiNiFi CPP. Also conan has really
> good support for cross platform compatibility for Linux and Windows. At this
> point because we already have most of MiNiFi CPP's external lib dependencies
> installed with conan, our CMake build will just focus on building the MiNiFi
> CPP code, enabling faster builds. Also I will need to account for backward
> compatibility support with the previous way of installing MiNiFi CPP external
> lib dependencies using CMake FetchContent_Declare(...) approach until
> conanfile.py approach covers installing all these dependencies. A valid
> verification of the MiNiFi C++ build would be to run GTESTs and if all the
> expected test cases pass, then it may be fair to conclude we would be ready
> to deprecate the slow CMake FetchContent_Declare(...) approach.
>
> *Steps to Perform MiNiFi C++ Build Enhancement (Plan to Integrate into
> script(s)):*
> # Install conan version 2 into MiNiFi C++ dev environment (my preference is
> using docker container)
> # Create a conanfile.py file at root of MiNiFi C++ project where we run
> "conan install" or "conan install --build=missing" to install MiNiFi C++
> external lib dependencies first. Then we run CMake to build MiNiFi C++.
> # Find all MiNiFi C++ find_package(...) CMake function calls and check if
> there is already supported conan packages on the conancenter, so we can
> install those packages.
> # Find all FetchContent_Declare(...) CMake function calls and check if there
> is already supported conan packages on the conancenter, so we can install
> those packages.
> # At this point we will have used conan version 2 C++ package manager to
> install almost all MiNiFi C++'s external dependencies.
> ## With Conan, we will have installed these MiNiFi external lib dependencies
> as prebuilt binary conan packages.
> ## Or we will have instructed conan to first build each C++ external lib
> dependency, upload each lib to conancenter.
> # Thus, conan manages installing MiNiFi C++'s external lib dependencies, so
> we can then run CMake to build MiNiFi faster.
>
> Here is my draft PR for this Jira Ticket for your reference:
> [https://github.com/apache/nifi-minifi-cpp/pull/1775]
> * I have added *abseil* external lib dependency to conanfile.py and
> integrated it into Abseil.cmake with conditional statement that checks if
> USE_CONAN_PACKAGER minifi cpp option is true else checks if
> USE_CMAKE_FETCH_CONTENT option is true for backward compatibility, so we can
> began the process of integrating in conan version 2.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
