Behavior of 'app/extract' as deployed by Genodians.org

[cnuke]

The way the extract component is implemented and operated in the genodians.org appliance can lead to undesired behavior. A few days ago updating the website stopped as one of the extract operations failed:

[…]
[init -> genodians -> import -> fetchurl] fetch https://github.com/nfeske/genodian/archive/master.zip to /download/nfeske.zip
[init -> genodians -> import -> fetchurl] fetch https://github.com/mickenx/genodian/archive/main.zip to /download/mickenx.zip
[init -> genodians -> import -> fetchurl] fetch https://github.com/m-stein/genodian/archive/master.zip to /download/m-stein.zip
[…]
[init -> genodians -> import -> extract] extracted '/download/m-stein.zip' to '/content/m-stein/'
[init -> genodians -> import -> extract] Warning: reading from archive /download/mickenx.zip failed
[init -> genodians -> import -> extract] Error: Uncaught exception of type 'Genode::Exception'
[init -> genodians -> import -> extract] Warning: abort called - thread: ep
[…]

Since the extract component aborted the managing sequence component dwelt in the extract step as it only reacts upon a child exiting.

[cnuke]

• 230d030 remedies the situation by allowing the user of the extract component to endorse suppressing of failures and always reports success while keeping the old behavior intact

• 5023577 changes the implementation to report the result via the exit value and to override the reported value but also removes the exception and alters the way the component behaves, e.g., it does not stop processing on failure

• 901a494 extends the former commit by stopping the processing again on the first failed extraction

Now, I am not overly happy with the commits implementation-wise but going by intend I favor the combination of the last ones. In particular for the gendians.org use-case where most of the time failures seem to be temporary processing the resources we can and trying the ones we could not in the next round worked well so far.

[nfeske]

Thanks @cnuke addressing this problem. Adding those new options looks sensible to me.

Implementation-wise, I think the bad smell comes from the use of exceptions. It would be very nice to follow up this work with the replacement of exceptions by the Attempt pattern. What do you think?

[cnuke]

@nfeske thanks for your suggestions, the following commits were adapted accordingly

• f19ced8 adds the README
• 129d3ba adds both options to influence the batch-processing
• cd39e3c picks up your remake about the exceptions and replaces them with the Attempt pattern.

[nfeske]

Beautiful work! Thank you very much. The README is the icing on the cake. :) Swiftly merged to staging.

[nfeske]

Merged in master.