python-pyarrow broken for parquet?

python-pyarrow broken for parquet?

[Phil Beadling]

Hi all,

When trying to use the parquet module in pyarrow - it's not finding the
internal module.  This is true for latest 4.0.1 and also the previous
version (shown below).

This is despite the fact that in the underlying apache-arrow package
parquet support is explicitly turned on as far as I can see:
https://github.com/guix-mirror/guix/blob/5ed105a8bb1a812975496dc3a091596355a0234c/gnu/packages/databases.scm#L3687

#:configure-flags
(list "-DARROW_PYTHON=ON"
"-DARROW_GLOG=ON"
;; Parquet options
"-DARROW_PARQUET=ON"
"-DPARQUET_BUILD_EXECUTABLES=ON".....


Any ideas how to resolve?

Thanks,
Phil.


$ guix environment --ad-hoc python python-pyarrow
The following derivation will be built:
   /gnu/store/jvc3sbqs2nssr8gdvvyd0hgf5z80ra5b-profile.drv

building CA certificate bundle...
listing Emacs sub-directories...
building fonts directory...
building directory of Info manuals...
building database for manual pages...
building profile with 2 packages...

$ python3
Python 3.8.2 (default, Jan  1 1970, 00:00:01)
[GCC 7.5.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
*>>> import pyarrow.parquet*
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File
"/gnu/store/53issxa0jw7l2wb7xbhp1jp5j7a91dd9-profile/lib/python3.8/site-packages/pyarrow/parquet.py",
line 33, in <module>
    import pyarrow._parquet as _parquet
*ModuleNotFoundError: No module named 'pyarrow._parquet'*
>>>

Re: python-pyarrow broken for parquet?

[zimoun]

Hi,

On Wed, 30 Jun 2021 at 21:10, Phil Beadling <phil@beadling.co.uk> wrote:

> When trying to use the parquet module in pyarrow - it's not finding the
> internal module.  This is true for latest 4.0.1 and also the previous
> version (shown below).
>
> This is despite the fact that in the underlying apache-arrow package
> parquet support is explicitly turned on as far as I can see:
> https://github.com/guix-mirror/guix/blob/5ed105a8bb1a812975496dc3a091596355a0234c/gnu/packages/databases.scm#L3687
>
> #:configure-flags
> (list "-DARROW_PYTHON=ON"
> "-DARROW_GLOG=ON"
> ;; Parquet options
> "-DARROW_PARQUET=ON"
> "-DPARQUET_BUILD_EXECUTABLES=ON".....

The package 'apache-arrow' is built with the Parquet support but not
the package 'python-pyarrow'.  To add the support of Parquet to the
Python binding, the environment variable PYTHON_WITH_PARQUET should be
set to 1, IIRC the doc.

Tweaking the 'python-pyarrow' with:

--8<---------------cut here---------------start------------->8---
      (add-after 'unpack 'set-env
        (lambda _ (setenv "PYARROW_WITH_PARQUET" "1") #t)))))
--8<---------------cut here---------------end--------------->8---

then running "./pre-inst-env guix build python-pyarrow --no-grafts" I get:

--8<---------------cut here---------------start------------->8---
-- Found Arrow:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Arrow version: 4.0.1 (CMake package configuration: Arrow)
-- Arrow SO and ABI version: 400
-- Arrow full SO version: 400.1.0
-- Found the Arrow core shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow.so
-- Found the Arrow core import library:
-- Found the Arrow core static library:
-- Could NOT find ArrowPython (missing: ArrowPython_DIR)
-- Checking for module 'arrow-python'
--   Found arrow-python, version 4.0.1
-- Found ArrowPython:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Found the Arrow Python by pkg-config: arrow-python
-- Found the Arrow Python shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow_python.so
-- Found the Arrow Python import library:
-- Found the Arrow Python static library: ARROW_PYTHON_static_lib-NOTFOUND
-- Could NOT find Parquet (missing: Parquet_DIR)
-- Checking for module 'parquet'
--   Found parquet, version 4.0.1
-- Found Parquet:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib//gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Parquet version: 4.0.1 (pkg-config: parquet)
-- Found the Parquet shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libparquet.so
-- Found the Parquet import library:
-- Found the Parquet static library: PARQUET_static_lib-NOTFOUND
-- Configuring done
CMake Error in CMakeLists.txt:
  Imported target "parquet_shared" includes non-existent path

    "/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib//gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include"

  in its INTERFACE_INCLUDE_DIRECTORIES.  Possible reasons include:

  * The path was deleted, renamed, or moved to another location.

  * An install or uninstall procedure did not complete successfully.

  * The installation package was faulty and references files it does not
  provide.


CMake Error in CMakeLists.txt:
  Imported target "parquet_shared" includes non-existent path

    "/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib//gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include"

  in its INTERFACE_INCLUDE_DIRECTORIES.  Possible reasons include:

  * The path was deleted, renamed, or moved to another location.

  * An install or uninstall procedure did not complete successfully.

  * The installation package was faulty and references files it does not
  provide.
--8<---------------cut here---------------end--------------->8---

Without this environment variable PYTHON_WITH_PYARROW set to 1, the
configuration is not looking after Parquet, for instance:

--8<---------------cut here---------------start------------->8---
-- Found Arrow:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Arrow version: 4.0.1 (CMake package configuration: Arrow)
-- Arrow SO and ABI version: 400
-- Arrow full SO version: 400.1.0
-- Found the Arrow core shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow.so
-- Found the Arrow core import library:
-- Found the Arrow core static library:
-- Could NOT find ArrowPython (missing: ArrowPython_DIR)
-- Checking for module 'arrow-python'
--   Found arrow-python, version 4.0.1
-- Found ArrowPython:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Found the Arrow Python by pkg-config: arrow-python
-- Found the Arrow Python shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow_python.so
-- Found the Arrow Python import library:
-- Found the Arrow Python static library: ARROW_PYTHON_static_lib-NOTFOUND
-- Configuring done
-- Generating done
-- Build files have been written to:
/tmp/guix-build-python-pyarrow-4.0.1.drv-0/source/python/build/temp.linux-x86_64-3.8
-- Finished cmake for pyarrow
-- Running cmake --build for pyarrow
cmake --build . --config release --
make[1]: Entering directory
'/tmp/guix-build-python-pyarrow-4.0.1.drv-0/source/python/build/temp.linux-x86_64-3.8'
--8<---------------cut here---------------end--------------->8---

Well, setting the environment variable is not enough.  :-)  Do you
want to give a try for fixing it?


All the best,
simon

Re: python-pyarrow broken for parquet?

[phil]

Thanks Simon.

Yep I got this far too - and I have a candidate fix for building parquet.  But it's tremendously hacky (sed'ing hardcoded variables into the cmake files to trample the derived settings in several places).  It seems to work but needs finessing.  I'll post here shortly, but not sure it's stable enough to be updated in Guix proper.  We can debate that when everyone sees my horrendous fix.



-----Original Message-----
From: zimoun <zimon.toutoune@gmail.com>
To: Phil Beadling <phil@beadling.co.uk>
Cc: help-guix <help-guix@gnu.org>
Sent: Fri, 02 Jul 2021 12:01
Subject: Re: python-pyarrow broken for parquet?

Hi,

On Wed, 30 Jun 2021 at 21:10, Phil Beadling <phil@beadling.co.uk> wrote:

> When trying to use the parquet module in pyarrow - it's not finding the
> internal module.  This is true for latest 4.0.1 and also the previous
> version (shown below).
>
> This is despite the fact that in the underlying apache-arrow package
> parquet support is explicitly turned on as far as I can see:
> https://github.com/guix-mirror/guix/blob/5ed105a8bb1a812975496dc3a091596355a0234c/gnu/packages/databases.scm#L3687
>
> #:configure-flags
> (list "-DARROW_PYTHON=ON"
> "-DARROW_GLOG=ON"
> ;; Parquet options
> "-DARROW_PARQUET=ON"
> "-DPARQUET_BUILD_EXECUTABLES=ON".....

The package 'apache-arrow' is built with the Parquet support but not
the package 'python-pyarrow'.  To add the support of Parquet to the
Python binding, the environment variable PYTHON_WITH_PARQUET should be
set to 1, IIRC the doc.

Tweaking the 'python-pyarrow' with:

--8<---------------cut here---------------start------------->8---
      (add-after 'unpack 'set-env
        (lambda _ (setenv "PYARROW_WITH_PARQUET" "1") #t)))))
--8<---------------cut here---------------end--------------->8---

then running "./pre-inst-env guix build python-pyarrow --no-grafts" I get:

--8<---------------cut here---------------start------------->8---
-- Found Arrow:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Arrow version: 4.0.1 (CMake package configuration: Arrow)
-- Arrow SO and ABI version: 400
-- Arrow full SO version: 400.1.0
-- Found the Arrow core shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow.so
-- Found the Arrow core import library:
-- Found the Arrow core static library:
-- Could NOT find ArrowPython (missing: ArrowPython_DIR)
-- Checking for module 'arrow-python'
--   Found arrow-python, version 4.0.1
-- Found ArrowPython:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Found the Arrow Python by pkg-config: arrow-python
-- Found the Arrow Python shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow_python.so
-- Found the Arrow Python import library:
-- Found the Arrow Python static library: ARROW_PYTHON_static_lib-NOTFOUND
-- Could NOT find Parquet (missing: Parquet_DIR)
-- Checking for module 'parquet'
--   Found parquet, version 4.0.1
-- Found Parquet:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib//gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Parquet version: 4.0.1 (pkg-config: parquet)
-- Found the Parquet shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libparquet.so
-- Found the Parquet import library:
-- Found the Parquet static library: PARQUET_static_lib-NOTFOUND
-- Configuring done
CMake Error in CMakeLists.txt:
  Imported target "parquet_shared" includes non-existent path

    "/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib//gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include"

  in its INTERFACE_INCLUDE_DIRECTORIES.  Possible reasons include:

  * The path was deleted, renamed, or moved to another location.

  * An install or uninstall procedure did not complete successfully.

  * The installation package was faulty and references files it does not
  provide.


CMake Error in CMakeLists.txt:
  Imported target "parquet_shared" includes non-existent path

    "/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib//gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include"

  in its INTERFACE_INCLUDE_DIRECTORIES.  Possible reasons include:

  * The path was deleted, renamed, or moved to another location.

  * An install or uninstall procedure did not complete successfully.

  * The installation package was faulty and references files it does not
  provide.
--8<---------------cut here---------------end--------------->8---

Without this environment variable PYTHON_WITH_PYARROW set to 1, the
configuration is not looking after Parquet, for instance:

--8<---------------cut here---------------start------------->8---
-- Found Arrow:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Arrow version: 4.0.1 (CMake package configuration: Arrow)
-- Arrow SO and ABI version: 400
-- Arrow full SO version: 400.1.0
-- Found the Arrow core shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow.so
-- Found the Arrow core import library:
-- Found the Arrow core static library:
-- Could NOT find ArrowPython (missing: ArrowPython_DIR)
-- Checking for module 'arrow-python'
--   Found arrow-python, version 4.0.1
-- Found ArrowPython:
/gnu/store/bx4lprdcq5zw372f2rcc27c412chximr-apache-arrow-4.0.1-include/share/include
(found version "4.0.1")
-- Found the Arrow Python by pkg-config: arrow-python
-- Found the Arrow Python shared library:
/gnu/store/cagr2n03155mdx1mim8cxva1v22qhh0y-apache-arrow-4.0.1-lib/lib/libarrow_python.so
-- Found the Arrow Python import library:
-- Found the Arrow Python static library: ARROW_PYTHON_static_lib-NOTFOUND
-- Configuring done
-- Generating done
-- Build files have been written to:
/tmp/guix-build-python-pyarrow-4.0.1.drv-0/source/python/build/temp.linux-x86_64-3.8
-- Finished cmake for pyarrow
-- Running cmake --build for pyarrow
cmake --build . --config release --
make[1]: Entering directory
'/tmp/guix-build-python-pyarrow-4.0.1.drv-0/source/python/build/temp.linux-x86_64-3.8'
--8<---------------cut here---------------end--------------->8---

Well, setting the environment variable is not enough.  :-)  Do you
want to give a try for fixing it?


All the best,
simon

Re: python-pyarrow broken for parquet?

[Phil Beadling]

As promised - this works for me but the patching of the make files, in
particular the 2 sed commands is very brittle to any changes in the
underlying project.  I'm not sure it should go into Guix proper as-is, but
if people think it's useful I'm happy to submit the patch.

I may try to improve on this when I have moment, but if someone else wants
to run with it - it's a good starting point at least.

The problem is that the generation of the PARQUET_INCLUDE_DIR nad
PARQUET_LIB_DIR end up concatentating both the include and lib dirs
together:

For example debugging the cmake file the value
"PARQUET_INCLUDE_DIR/parquet" becomes:
"*/gnu/store/ywklhws3ccb457gsb605z95azbfpsbyl-apache-arrow-3.0.0-lib/*/gnu/store/zzzb4ymfj3igynsflxwxsn58kvnpa6qb-apache-arrow-3.0.0-include/share/include/parquet"


The *lib *directory shouldn't be there at all.

--8<---------------cut here---------------start------------->8---
(define-public python-pyarrow-parquet
  (package/inherit python-pyarrow
                   (arguments
                    (substitute-keyword-arguments (package-arguments
python-pyarrow)
                      ((#:phases phases)
                       `(modify-phases ,phases
                          (add-before 'install 'patch-cmake-variables
                            (lambda* (#:key inputs #:allow-other-keys)
                              ;; Replace cmake locations with hardcoded
guix links for the underlying C++ lib - this is a pretty awful hack
                              (invoke "sed" "-i" (string-append
"1s#^#set(PARQUET_INCLUDE_DIR \"" (assoc-ref inputs "apache-arrow:include")
"/share/include\
\")\\n#") "cmake_modules/FindParquet.cmake")
                              (invoke "sed" "-i" (string-append
"116s#^#set(PARQUET_LIB_DIR \"" (assoc-ref inputs "apache-arrow:lib")
"/lib\")\\n#") "cmake\
_modules/FindParquet.cmake")))
                          (add-before 'install 'patch-parquet-library
                            (lambda _
                              ;; Another nasty hack - there must be a
better way to change this?
                              (substitute* "CMakeLists.txt"
(("parquet_shared") "parquet"))))
                          (add-before 'install 'set-PYARROW_WITH_PARQUET
                            (lambda _
                              (setenv "PYARROW_WITH_PARQUET" "1")
                              ;;(setenv "VERBOSE" "1") ;; useful debug for
cmake
                              #t))))))
                   ;; we need includes from apache as well as libs for
parquet
                   (propagated-inputs
                    `(("python-pandas" ,python-pandas-simm)
                      ("apache-arrow:lib" ,apache-arrow "lib")
                      ("apache-arrow:include" ,apache-arrow "include")
                      ,@(fold alist-delete (package-propagated-inputs
python-pyarrow)
                              '("python-pandas" "apache-arrow"))))))
--8<---------------cut here---------------end--------------->8---

On Fri, 2 Jul 2021 at 16:34, <phil@beadling.co.uk> wrote:

> Thanks Simon.
>
> Yep I got this far too - and I have a candidate fix for building parquet.
> But it's tremendously hacky (sed'ing hardcoded variables into the cmake
> files to trample the derived settings in several places).  It seems to work
> but needs finessing.  I'll post here shortly, but not sure it's stable
> enough to be updated in Guix proper.  We can debate that when everyone sees
> my horrendous fix.
>
>
>

Re: python-pyarrow broken for parquet?

[Phil Beadling]

Apologies one tiny correction - the version of python-pandas probably
doesn't need changing - in the package I update to a more recent custom
build of pandas, I'm using - but it's probably not necessary:
("python-pandas" ,python-pandas-simm)

On Mon, 5 Jul 2021 at 13:13, Phil Beadling <phil@beadling.co.uk> wrote:

> As promised - this works for me but the patching of the make files, in
> particular the 2 sed commands is very brittle to any changes in the
> underlying project.  I'm not sure it should go into Guix proper as-is, but
> if people think it's useful I'm happy to submit the patch.
>
> I may try to improve on this when I have moment, but if someone else wants
> to run with it - it's a good starting point at least.
>
> The problem is that the generation of the PARQUET_INCLUDE_DIR nad
> PARQUET_LIB_DIR end up concatentating both the include and lib dirs
> together:
>
> For example debugging the cmake file the value
> "PARQUET_INCLUDE_DIR/parquet" becomes:
> "*/gnu/store/ywklhws3ccb457gsb605z95azbfpsbyl-apache-arrow-3.0.0-lib/*/gnu/store/zzzb4ymfj3igynsflxwxsn58kvnpa6qb-apache-arrow-3.0.0-include/share/include/parquet"
>
>
> The *lib *directory shouldn't be there at all.
>
> --8<---------------cut here---------------start------------->8---
> (define-public python-pyarrow-parquet
>   (package/inherit python-pyarrow
>                    (arguments
>                     (substitute-keyword-arguments (package-arguments
> python-pyarrow)
>                       ((#:phases phases)
>                        `(modify-phases ,phases
>                           (add-before 'install 'patch-cmake-variables
>                             (lambda* (#:key inputs #:allow-other-keys)
>                               ;; Replace cmake locations with hardcoded
> guix links for the underlying C++ lib - this is a pretty awful hack
>                               (invoke "sed" "-i" (string-append
> "1s#^#set(PARQUET_INCLUDE_DIR \"" (assoc-ref inputs "apache-arrow:include")
> "/share/include\
> \")\\n#") "cmake_modules/FindParquet.cmake")
>                               (invoke "sed" "-i" (string-append
> "116s#^#set(PARQUET_LIB_DIR \"" (assoc-ref inputs "apache-arrow:lib")
> "/lib\")\\n#") "cmake\
> _modules/FindParquet.cmake")))
>                           (add-before 'install 'patch-parquet-library
>                             (lambda _
>                               ;; Another nasty hack - there must be a
> better way to change this?
>                               (substitute* "CMakeLists.txt"
> (("parquet_shared") "parquet"))))
>                           (add-before 'install 'set-PYARROW_WITH_PARQUET
>                             (lambda _
>                               (setenv "PYARROW_WITH_PARQUET" "1")
>                               ;;(setenv "VERBOSE" "1") ;; useful debug for
> cmake
>                               #t))))))
>                    ;; we need includes from apache as well as libs for
> parquet
>                    (propagated-inputs
>                     `(("python-pandas" ,python-pandas-simm)
>                       ("apache-arrow:lib" ,apache-arrow "lib")
>                       ("apache-arrow:include" ,apache-arrow "include")
>                       ,@(fold alist-delete (package-propagated-inputs
> python-pyarrow)
>                               '("python-pandas" "apache-arrow"))))))
> --8<---------------cut here---------------end--------------->8---
>
> On Fri, 2 Jul 2021 at 16:34, <phil@beadling.co.uk> wrote:
>
>> Thanks Simon.
>>
>> Yep I got this far too - and I have a candidate fix for building
>> parquet.  But it's tremendously hacky (sed'ing hardcoded variables into the
>> cmake files to trample the derived settings in several places).  It seems
>> to work but needs finessing.  I'll post here shortly, but not sure it's
>> stable enough to be updated in Guix proper.  We can debate that when
>> everyone sees my horrendous fix.
>>
>>
>>

Re: python-pyarrow broken for parquet?

[Ricardo Wurmus]

Hi Phil,

> Apologies one tiny correction - the version of python-pandas 
> probably
> doesn't need changing - in the package I update to a more recent 
> custom
> build of pandas, I'm using - but it's probably not necessary:
> ("python-pandas" ,python-pandas-simm)

I’m working on upgrading Pandas, and I’ve adapted your changes 
(not using “sed” but a substitute* form) to fix pyarrow.  I’ll 
push the changes soon.

Thank you!

-- 
Ricardo

Re: python-pyarrow broken for parquet?

[Phil]

Ricardo Wurmus writes:

> I’m working on upgrading Pandas, and I’ve adapted your changes (not
> using “sed” but a substitute* form) to fix pyarrow.  I’ll push the
> changes soon.

Great - thanks very much for including Ricardo!

Re: python-pyarrow broken for parquet?

[zimoun]

Hi,

On Wed, 28 Jul 2021 at 21:37, Phil <phil@beadling.co.uk> wrote:
> Ricardo Wurmus writes:
>
>> I’m working on upgrading Pandas, and I’ve adapted your changes (not
>> using “sed” but a substitute* form) to fix pyarrow.  I’ll push the
>> changes soon.
>
> Great - thanks very much for including Ricardo!

Thanks for your work. :-)
And thanks Ricardo for the fix.

All the best,
simon