Fix font problems in devteam/xy_plot tool

[jennaj]

MTS: https://toolshed.g2.bx.psu.edu/view/devteam/xy_plot/ecb437f1d298

Tool: Plotting Tool for multiple series and graph types | 1.0.2

Used in this GTN tutorial, section "Hands-on: Plot Rarefaction": http://galaxyproject.github.io/training-material/topics/metagenomics/tutorials/mothur-miseq-sop/tutorial.html

[jennaj]

ping @davebx

[martenson]

updated on Main in 34a4dd4

[jennaj]

Retesting

[jennaj]

@martenson @natefoo version 1.0.2 fails.

Does this look like a Galaxy config problem or a tool problem?

Dataset Error
An error occured while running the tool toolshed.g2.bx.psu.edu/repos/devteam/xy_plot/XY_Plot_1/1.0.2.

Tool execution generated the following messages:

Fontconfig warning: "/etc/fonts/fonts.conf", line 146: blank doesn't take any effect anymore. please remove it from your fonts.conf

[martenson]

@jennaj can you please share the history?

edit: found it in the bug report

from @nekrut:

We should remove this tool. It was a proof of concept tool James wrote during an Encode meeting in 2008. New charts can do the same thing. a.

[martenson]

I propose hiding the tool and adjusting the tutorial to use charts.

[natefoo]

This can be fixed relatively easily by using exit codes instead of stderr.

The warning is caused by conda using a newer version of fontconfig than the one on the system, for which /etc/fonts/fonts.conf is written. There's a suitable <env>/etc/fonts/fonts.conf in the conda env but I can't think of a way to get $FONTCONFIG_FILE set properly for this.

[martenson]

@natefoo the actual error is different (can be seen in the history: https://usegalaxy.org/histories/view?id=223db61e77830a8e).

Error in Summary.factor(c(1L, 2L, 49L, 50L, 51L, 52L, 53L, 54L, 55L, 56L,  : 
  ‘range’ not meaningful for factors

[natefoo]

That's hid 516, hid 517 was successful though.

[martenson]

Oh so you think 516 had bad inputs? 517 is definitely just warning written to stderr, I can see the plot fine.

[natefoo]

Well, interesting development... fontconfig in the R conda env is correctly reading its own fonts.conf. But when we run jobs, we set $FONTCONFIG_FILE to /cvmfs/main.galaxyproject.org/fontconfig/fonts.conf, which loads some custom fonts in /cvmfs/main.galaxyproject.org/fontconfig/fonts and then includes /etc/fonts/fonts.conf. This was done long, long ago for pre-TS tools.

If I don't set $FONTCONFIG_FILE, it'll probably fix this tool, but I don't know what tool(s) used these fonts and whether they will break if these fonts aren't available. If I knew what tools they were, I could set $FONTCONFIG_FILE just for these tools.

[natefoo]

However, I am inclined to just unset this and fix any old tools it breaks as they are discovered. Scatterplot and Histogram are the likely subjects.

[natefoo]

Well this is interesting. I ran Plotting tool, Scatterplot, and Histogram, then disabled $FONTCONFIG_FILE and ran all 3 again. With $FONTCONFIG_FILE set, Plotting tool ends in error but produces a valid plot:

With it unset, Plotting tool ends ok but the fonts in the plot are symbol fonts:

Unsetting it had no effect on Scatterplot or Histogram.

I think maybe the tool should add a requirement on some font package in conda?

[jennaj]

test history (late, you found it in other ticket): https://usegalaxy.org/u/jen/h/16s-microbial-analysis-with-mothur---proccess

Dataset 517 is in an error state (red). The graph does look how I would expect it to (but can't really double check, am comparing with whatever tool version was used to create tutorial graphics, this has never been on main before and haven't checked at eu).

[jennaj]

@natefoo right, that was the problem before - red error dataset but graphics produced. If you create a ticket for a tool update, could you link it back? That might be a less risky choice (adding conda dep to tool) rather than making a change that impacts all tools.

[jennaj]

yes, hid 516 failed because of usage - The way the tool form is designed, each series input needs to have the "Header in first line?" set to yes individually. This is easy to miss, especially when doing that step in the tutorial where you add in 10+ series, each its own section. The form could be modified or the stderr could be more informative (if can't parse the first line, fail but warn user "Do the inputs have headers? Currently set as "No" for series N."

But considering how much other stuff is going on, is minor, so didn't report it, just the hid 517 failure. I'll deal with that later, especially if users start to report running into it (expect they will, students/new users use these tutorials).

[natefoo]

I've fixed this by installing DejaVu fonts into CVMFS and making fontconfig aware of them with $XDG_DATA_HOME. The process was:

1. cvmfs_server transaction $repo && mkdir -p /cvmfs/$repo/xdg/data/fonts
2. curl -L http://sourceforge.net/projects/dejavu/files/dejavu/2.37/dejavu-fonts-ttf-2.37.tar.bz2 | tar jxf - -C /cvmfs/$repo/xdg/data/fonts
3. Unset $FONTCONFIG_FILE and set XDG_DATA_HOME=/cvmfs/$repo/xdg/data in 502a0b3

(where $repo is test.galaxyproject.org or main.galaxyproject.org)

[jennaj]

thanks for testing, also worked for me. changing test status.