Problem when moving _targets from Unix to Windows

[shambhu112]

Issue

I have a targets repository folder I have created on ubuntu . When I export the rstudio folder project as a zip and run this folder in windows I get the following error : Error in gzfile(file, "rb") : invalid 'description' argument This error happens at the time of

targets::tar_read_raw("bank_2020_kpis") I have tried targets::tar_load("bank_2020_kpis") also.

Is this is a known issue between unix and windows ?

sessionInfo:= on ubuntu
R version 4.0.2 (2020-06-22) Platform: x86_64-pc-linux-gnu (64-bit) Running under: Ubuntu 20.04.2 LTS Matrix products: default BLAS: /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.9.0 LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.9.0

I can update this issue with a recreatable example. It will take me a few days

[wlandau]

It's hard to say at the moment, but here are some possible explanations:

• The files _targets/objects/bank_2020_kpis and/or _targets/meta/meta may be missing.
• You might have different R versions or package versions on Ubuntu vs Windows.
• You may be running R in a working directory other than the project root. In order to read the target, list.files(getwd()) must include "_targets".

The error "Error in gzfile(file, "rb") : invalid 'description' argument" usually comes up during readRDS() when some kind of output is missing or invalid. So one thing you could try is readRDS("_targets/objects/bank_2020_kpis") on both machines.

[shambhu112]

Thanks for your quick reply.
a few updates

1. both '_targets/objects/bank_2020_kpi' and '_targets/meta/meta' exists
2. Running R in the right Working Directory
3. readRDS("_targets/objects/bank_2020_kpis") works well.

Let me dig into this more and get back. At this time, I will probably do the workaround of 'readRDS("_targets/objects/bank_2020_kpis")' and move forward

[shirdekel]

I'm also having this issue between two Macs (both 10.14.6). The files are shared between them using Dropbox.

The call to targets::tar_read(object_name) works for one computer, but for the other I get Error in gzfile(file, "rb") : invalid 'description' argument.

For both computers:

• readRDS("_targets/objects/object_name") works
• _targets/meta/meta exists
• list.files(getwd()) includes "_targets"
• targets::tar_make() runs fine

[wlandau]

Thanks for the example. I ran tar_make() on https://github.com/shirdekel/targets-test/blob/master/_targets.R using 0.2.0, then successively again on 0.2.0, 0.3.0, 0.3.1, 0.4.1, 0.4.2, and 0.5.0.9002. After the first tar_make(), each successive tar_make() correctly skipped the target. So maybe it has something to do with other parts of your package environment? What if you used an renv library to hold all your packages constant across computers? And what about R versions?

The metadata you mention changed because of #407, but I thought I did that compatibly. (Looks that way locally on my machine at any rate.)

What does tar_sitrep() say when the target is incorrectly invalidated? What cues are activated?

[shirdekel]

The way I reproduce it is by installing version 0.2.0 on Computer 2, running tar_make() on Computer 1 (which has version, 0.4.2.9000) and then calling tar_read(test) on Computer 2.

I assumed it was a targets issue because I didn't use renv to reproduce this and R versions stayed constant.

tar_sitrep() gives the following in Computer 2 after I see the error:

# A tibble: 1 x 9
  name  record always never command depend format iteration file 
  <chr> <lgl>  <lgl>  <lgl> <lgl>   <lgl>  <lgl>  <lgl>     <lgl>
1 test  FALSE  FALSE  FALSE FALSE   FALSE  FALSE  FALSE     TRUE 

[shirdekel]

But actually, I'm struggling to reproduce it if I do the same procedure within one computer. So you might be right about it being something else.

[wlandau]

The way I reproduce it is by installing version 0.2.0 on Computer 2, running tar_make() on Computer 1 (which has version, 0.4.2.9000) and then calling tar_read(test) on Computer 2.

That actually makes sense. If you run the pipeline with targets >= 0.4.2, then targets <= 0.3.1 will not be able to read the data because of #407. Upgrading targets on Computer 2 should solve that the issue.

As of #407, the metadata no longer records file paths unless you use format = "file" or format = "aws_file" in tar_target() or tar_option_set(). That's why 0.2.0 cannot read from newer data stores.

[shirdekel]

Ah ok. Thanks for looking into it!