Max Nikulin <maniku...@gmail.com> writes:
>> #!/bin/bash
>> gjl_package=plantuml
>> gjl_jar="plantuml.jar"
>> source /usr/share/java-config-2/launcher/launcher.bash
>
> It seems with with the suggested patch both -Djava.awt.headless=true and
> -headless are passed to jar. I suspect it is either redundant or
> launcher.bash converts "-headless" to "-Djava.awt.headless=true". In
> ubuntu there is no room for special treatment of "-headless" option by
> the shell wrapper:
It is redundant, but should not cause any harm. plantuml source code has:
if (argsArray.length > 0 && argsArray[0].equalsIgnoreCase("-headless")) {
System.setProperty("java.awt.headless", "true");
}
I can of course add a condition to not add -headless if we are running
jar, but I do not see much benefit except purity argument.
>> I feel that running headless mode in the ob-plantuml is deliberate.
>
> https://plantuml.com/faq
> "Basically, this is because PlantUML needs to have access to some
> graphical resources"
>
> I have no idea what is actually affected.
It mostly looks like otherwise there are some harmless warnings in the
output.
> Which peace of code (java or plantuml.jar) handles "-headless" option?
> This question affects decision if jar and executable wrapper modes may
> share same list of arguments. Plantuml site recommends namely
> -Djava.awt.headless=true. Anyway with or without the patch,
> "-Djava.awt.headless=true" is added unconditionally in the jar mode. I
> would revert that patch and either add it to java header argument or to
> a new defcustom.
I do not see much point customising this particular argument. We _do
not_ want plantuml in gui mode. Ever. Otherwise, babel behaviour will be
unpredictable. If we make it customizeable, the only "benefit" would be
allowing users to shoot their own leg.
> Are there cases when :cmdline header argument is not enough and extra
> custom variable org-plantuml-args (org-plantump-executable-args) should
> be used?
> Actually I have realized that I do not understand why babel code uses
> mix of defcustom and header arguments. The latter are more flexible and
> may be adjusted per source block. On the other hand they may be set in
> init file. Even java executable and path to jar file may be obtained
> from header arguments.
Good questions. I think that the main reason is that most of header args
are not merged - only the last value is used. If you define :cmdline in
source block, it will override system-wide settings. Having an
additional defcustom allows to provide a "permanent" set of parameters -
to be used no matter what.
I realise that the current situation is kind of awkward. Maybe we should
encourage ob-* libraries to use special header args instead of
defcustoms and modify built-in libraries in this spirit. Would you be
interested to write a feature request?
Best,
Ihor
