Closed MarekGierlinski closed 3 months ago
Thank you, I can reproduce. Looks like there is a change in R 4.4.
Could you try the latest commit? This was related to the change in how R handled promises. The solution was to force evaluation of promises before serialization.
Yes, the latest commit works well. Thank you. Do you know how long until it is in CRAN?
Just to add to my original post, in a different project using targets with tar_option_set(format = "qs")
, while using qs
0.26.1 I have encountered the following crash:
*** caught segfault ***
address 0x0, cause 'memory not mapped'
/
Traceback:
1: qs::qsave(x = object, file = path, preset = preset)
2: store_write_path.tar_qs(store, store_convert_object(store, object), stage)
3: store_write_path(store, store_convert_object(store, object), stage)
4: store_write_object.default(target$store, target$value$object)
5: store_write_object(target$store, target$value$object)
6: builder_update_object(target)
It seemed to appear at random, for different targets, and I could not reproduce it in a smaller example. See also this discussion.
I'm happy to report that your latest GitHub release (0.26.3) solved the problem. So, whatever whatever you did to rectify my original issue, helped with this one too.
Out on CRAN now
I have come across an error involving
cowplot::plot_grid()
andqs::qsave()
. I usetargets
with optionformat = "qs"
and after recent updates some of my pipelines started to experience crashes. I tried using older version ofqs
andcowplot
, but the error persisted, so I suspect it is happened after upgrading R to 4.4.0.Here is a reproducible (I hope) example showing when the error happens. The code below creates two
ggplot2
objects with nearly identical images. One is created withcowplot::grid_plot()
with other withpatchwork
. The object created bypatchwork
is saved with no issues, but saving the object created withcowplot
crashes with "bad binding access".Here is my session info: