Am Donnerstag, den 07.11.2013, 00:25 -0600 schrieb John Chilton:
My two cents below.
On Wed, Nov 6, 2013 at 4:20 PM, Björn Grüning <bjoern.gruening@pharmazie.uni-freiburg.de> wrote:
Hi Dave,
We're thinking that the following approach makes the most sense:
<action type="setup_perl_environment"> OR <action type="setup_r_environment"> OR <action type="setup_ruby_environment"> OR <action type="setup_virtualenv"> <repository changeset_revision="978287122b91" name="package_perl_5_18" owner="iuc" toolshed="http://testtoolshed.g2.bx.psu.edu"> <package name="perl" version="5.18.1" /> </repository> <repository changeset_revision="8fc96166cddd" name="package_expat_2_1" owner="iuc" toolshed="http://testtoolshed.g2.bx.psu.edu"> <package name="expat" version="2.1" /> </repository> </action>
For all <repository> tag sets contained within these setup_* tags, the repository's env.sh would be pulled in for the setup of the specified environment without requiring a set_environment_for_install action type.
Would this work for your use cases?
Yes, the first one. But its a little bit to verbose or? Include the perl repository in a setup_perl environment should be implicit or? We can assume that this need to be present. Do you have example why sourcing every <repository> by default can be harmful? It would make such an installation so much easier and less complex.
I am not sure I understand this paragraph - I have a vague sense I agree but is there any chance you could rephrase this or elaborate?
My first use case will be addressed by this suggestion. I had hoped that we can create a less verbose syntax. If we I specify a package at the top of my xml file: <package name="expat" version="2.1.0"> <repository name="package_expat_2_1" owner="iuc" prior_installation_required="True" /> </package> I need to repeat it either in a <action type="set_environment_for_install"> or in a <action type="setup_perl_environment">. My hope was to get rid of these. Once a package definition is specified/build, every ENV var is available in any downstream <package>. But if there is any downsides or pitfalls this more verbose and explicit syntax will work for my usecase.
Also that did not solve the second use case. If have two <packages> one that is installing perl libraries and the second a binary that is checking or that needs these perl libs.
We have discussed off list in another thread. Just to summarize my thoughts there - I think we should delay this or not make it a priority if there are marginally acceptable workarounds that can be found for the time being. Getting these four actions to work well as sort of terminal endpoints and allow specification as tersely as possible should be the primary goal for the time being. You will see Perl or Python packages depend on C libraries 10 times more frequently than you will find makefiles and C programs depend on complex perl or python environments (correct me if I am wrong). Given that there is already years worth of tool shed development outlined in existing Trello cards - this is just how I would prioritize things (happy to be overruled).
Ok point taken. Lets focus on real issue. That use case is just a simplification / more structured way to write tool depdendecies, its not strictly needed to get my packages done. John just to make that use case clearer: - You have a package (A) with dependency (B) - B is not worth to put it in a extra repository (extra tool_dependencies.xml file) Currently, you are forced to define both in one <package> tag, because if you define it in two <package> tags A will not see B. The perl and python was a bad example you have that problem with every dependency that are not worth to put it in a separate repository. To summarize: I'm fine with that approach. It will address my current use case and it would be great to have it as proposed by Dave! Thanks a lot! Bjoern
If so, can you confirm that this should be done for all four currently supported setup_* action types?
I think it would be best to tackle setup_r_environment and setup_ruby_environment first. setup_virtualenv cannot have nested elements at this time - it is just assumed to be a bunch of text (either a file containing the dependencies or a list of the dependencies).
So setup_r_environment and setup_ruby_environment have the same structure:
<setup_ruby_environment> <repository .. /> <package .. /> <package .. /> </setup_ruby_environment>
... but setup_virtualenv is just
<setup_virtualenv>requests=1.20 pycurl==1.3</setup_virtualenv>
I have created a Trello card for this: https://trello.com/c/NsLJv9la (and some other related stuff).
Once that is tackled though, it will make sense to allow setup_virtualenv to utilize the same functionality.
Thanks all, -John
I think it will solve my current issues.
Based on your response, Greg or I will implement this as soon as possible.
Thanks! Bjoern
--Dave B.
On 11/06/2013 03:05 AM, Björn Grüning wrote:
Hi John,
Perl complicates things, TPP complicates things greatly.
So true, so true ...
Bjoern, can I ask you if this hypothetical exhibits the same problem and can be used to reason about these things more easily and drive a test implementation.
Yes to both questions :)
So right now, Galaxy has setup_virtualenv which will build and install Python packages in a virtual environment. However, some Python packages have C library dependencies that could prevent them from being installed.
As a specific example - take PyTables (install via "pip install tables") - which is a package for managing hierarchical datasets. If you try to install this with pip the way Galaxy will - it will fail if you do not have libhdf5 installed. So at a high-level, it would be nice if the tool shed had a libhdf5 definition and the dependencies file had some mechanism for declaring libhdf5 should be installed before a setup_virtualenv containing "tables" and its environment configured properly so the pip install succeeds (maybe just LD_LIBRARY_PATH needs to be set).
Indeed, same problem. I think we have this problem in every high-level install methodm because <set_environment_for_install> is not allowed as first action tag.
Can you think of any case where ENV vars can conflict with each other, besides set_to, and assuming that we source every env.sh file by default for every specified <package>.
Cheers, Bjoern
-John
On Tue, Nov 5, 2013 at 3:35 PM, Björn Grüning <bjoern.gruening@pharmazie.uni-freiburg.de> wrote:
Hi Greg,
Hello Bjoern,
On Nov 5, 2013, at 12:13 PM, Bjoern Gruening <bjoern.gruening@gmail.com> wrote:
> Hi Greg, > > I'm right now in implementing a setup_perl_environment and stumbled about > a tricky problem (that is not only related to perl but also for ruby, python > and R). > > The Problem: > Lets assume a perl package (A) requires a xml parser written in C/C++ (Z). > (Z) is a dependency that I can import but as far as I can see there is no > way to call set_environment_for_install before setup_perl_environment, > because setup_perl_environment defines an installation type.
The above is fairly difficult to understand - can you provide an actual xml recipe that provides some context?
Attached, please see a detailed explanation at the bottom.
> > I would like to discuss that issue to get a few ideas. I can think about > these solutions: > > - hackish solution: > I can call install_environment.add_env_shell_file_paths( action_dict[ > 'env_shell_file_paths' ] ) inside of the setup_*_environment path and remove > it from action type afterwards
Again, it's difficult to provide good feedback on the above approach without an example recipe. However, your "hackish solution" term probably means it is not ideal. ;)
:)
> > - import all env.sh variables from every (package) definition. Regardless > if set_environment_for_install is set or not.
I don't think the above approach would be ideal. It seems that it could fairly easily create conflicting environment variables within a single installation, so the latest value for an environment variable may not be what is expected.
What means conflicting ENV vars, I only can imagine multiple set_to that overwrite each other. append_to and prepend_to should be save or?
> I must admit, I do not understand why set_environment_for_install is > actually needed. I think we can assume that if I specify a > > <package name="R_3_0_1" version="3.0.1"> > <repository name="package_r_3_0_1" owner="iuc" > prior_installation_required="True" /> > </package> > > I want the ENV vars sourced.
Hmmm…so you are saying that you want the be able to define the above <package> tag set inside of an <actions> tag set and have everything work?
Oh no, I mean just have it as package like it is but source the env.sh file for every other <package> set automatically. So you do not need <set_environment_for_install>.
In the attached example: <package name="expat" version="2.1.0"> <repository changeset_revision="8fc96166cddd" name="package_expat_2_1" owner="iuc" prior_installation_required="True" toolshed="http://testtoolshed.g2.bx.psu.edu" /> </package>
Is not imported with <set_environment_for_install> so its actually useless (now). But the env.sh needs to be sourced during the "setup_perl_environment" part.
I think this may cause problems because I believe the <set_environment_for_install> tag set restricts activity to only the time when a dependent repository will be using the defined environment from the required repository in order to compile one or more of it's dependencies. Eliminating this restriction may cause problems after compilation. ALthough I cannot state this as a definite fact.
> Furthermore, that can solve an other issue: Namely, the need of ENV vars > from a package definition in the same file. Lets imagine package P has > dependency D and you want to download compile both in one > tool_dependencies.xml file. > You can either do it in one <package> definition or you need to split them > up in 2 tool_dependencies.xml files, rigth? > Maybe we can just assume a strict order in a tool_dependencies.xml file, > where every ENV vars are sourced for the following one? Does that make > sense?
It may make sense, but without an example it's diffiecult to answer this for sure. Can you provide some xml recipes that use your different proposals?
Sure, attached. Its quite complicated.
- TPP needs libgd to compile C-Code (no problem). - TPP needs some perl libs and perl -> "setup_perl_environment" (at runtime) - no problem until one of these perl packages (here XML-Parser) needs a C library (expat) - I don't see how to source expat during "setup_perl_environment" - TPP needs perl (at compile time) ... It would be more readable or logical to separate this recipe into two parts: TPP and Perl libraries
Something like that: <package name="trans_proteomic_pipeline_perl_libs" version="4.6.3"> ..... set PERL5LIBS </package> <package name="trans_proteomic_pipeline" version="4.6.3"> .... Here I need the PERL5LIBS </package>
- I don't see any way to get the PERL5LIBS from the perl libraries into a separate <package> section which tries to compile TPP.
> Any other ideas?
Not yet, but possibly after your next response. ;)
:) Here we go! Thanks Greg! Bjoern
> Thanks, > Bjoern
Thanks!
Greg Von Kuster
___________________________________________________________ Please keep all replies on the list by using "reply all" in your mail client. To manage your subscriptions to this and other Galaxy lists, please use the interface at: http://lists.bx.psu.edu/
To search Galaxy mailing lists use the unified search at: http://galaxyproject.org/search/mailinglists/
___________________________________________________________ Please keep all replies on the list by using "reply all" in your mail client. To manage your subscriptions to this and other Galaxy lists, please use the interface at: http://lists.bx.psu.edu/
To search Galaxy mailing lists use the unified search at: http://galaxyproject.org/search/mailinglists/
___________________________________________________________ Please keep all replies on the list by using "reply all" in your mail client. To manage your subscriptions to this and other Galaxy lists, please use the interface at: http://lists.bx.psu.edu/
To search Galaxy mailing lists use the unified search at: http://galaxyproject.org/search/mailinglists/