ete3 build fails on 64bit Debian/Ubuntu because of t-coffee

[marco-mariotti]

Hi, I've been struggling to run ete3 build in my Ubuntu 20.04.3 64bit machine. I finally solved the problem and I share the solution here.

The problem is that, after installing ete3 with conda (conda install -c etetoolkit ete3 ete_toolchain), when running even just:
ete3 build check
The command hangs, and a process (t_coffee) raises in memory usage until it is all eaten up.

This is due to a known issue of t_coffee related to MAX_N_PID. There is a workaround, however, which is to wrap t_coffee to set an env variable (MAX_N_PID_4_TCOFFEE) to prevent the problem: see implementation here (manual fix) and here (included in bioconda recipe for t-coffee).

Manual fix for those with the same issue:

1. download precompiled t_coffee (or install it from source) and set the workaround, following instructions in this comment
2. replace the ete_toolchain t_coffee with a link to the now functional t_coffee. For example, if you're in miniconda and your environment called ete3, and your functional t_coffee is ~/.t_coffee/bin/linux/t_coffee, this will do:
   mv ~/miniconda3/envs/ete3/bin/ete3_apps/bin/t_coffee ~/miniconda3/envs/ete3/bin/ete3_apps/bin/t_coffee.bug && ln -s ~/.t_coffee/bin/linux/t_coffee ~/miniconda3/envs/ete3/bin/ete3_apps/bin/t_coffee

In all cases, ete3 should implement a long term solution. In that regard, some useful information below.

There are two (bio)conda recipes for t_coffee that I know of, called "t_coffee", currently linking to version 11, and "t-coffee", linking to version 13. Only the second one has the workaround for the issue above. Ete_toolchain currently installs version 11 (perhaps it uses bioconda "t_coffee" recipe?). Unfortunately, the fix described above can't work at all for version 11, since t_coffee did not read MAX_N_PID_4_TCOFFEE back then.

So ete3 should install t_coffee version 13. While an apparently straightforward solution would seem to use the bioconda "t-coffee" recipe, this is a dependencies-heavy recipe (it's for the whole t_coffee "universe"), and has conflicts with other ete3 apps. So perhaps it is easier to adapt the current procedure used in ete_toolchain to install a newer (but light) t_coffee, then implement the workaround in the recipe like bioconda does.

[jbibbe4]

I just encountered the same problem when running ete3 build check after installing through conda.

The t-coffee recipe installs fine (no conflicts) in the same conda environment as ete3 (although I'm not sure how to check for conflicts if the dependencies of the other ete3 apps aren't installed/checked through conda)