Catch invalid device resolution or units (segmentation fault)

[hsbadr]

If the user passes large resolution or inappropriate units, some devices are unable to start or cause a segmentation fault.

Examples:

> ragg::agg_png(res = 100, units = "in")
Error in ragg::agg_png(res = 100, units = "in") :
  C++ exception: std::bad_alloc

> ragg::agg_png(res = 300, units = "in")

 *** caught segfault ***
address 0x14699b355000, cause 'invalid permissions'

Possible actions:
1: abort (with core dump, if enabled)
2: normal R exit
3: exit R without saving workspace
4: exit R saving workspace
Selection:

This is caught by grDevices as follows:

> grDevices::png(res = 300, units = "in")
Error in grDevices::png(res = 300, units = "in") :
  unable to start device 'png'
In addition: Warning message:
In grDevices::png(res = 300, units = "in") :
  cairo error 'invalid value (typically too big) for the size of the input (surface, pattern, etc.)'

[thomasp85]

This is probably a good idea, thanks!

[thomasp85]

This will now give:

> ragg::agg_png(res = 100, units = "in")
Fejl i ragg::agg_png(res = 100, units = "in") : 
  Memory allocation error. You are likely trying to create too large an image

The second example is harder to handle correctly because the OS might not always throw an allocation error that can be handled and the program will thus resume until it tries to access the non-allocated memory (more detail here: https://stackoverflow.com/a/9456758/987769)

[thomasp85]

I've decided to set a maximum dimension which is controllable through an option

[hsbadr]

This will now give:

Thanks @thomasp85 !

The second example is harder to handle correctly

I see.

I've decided to set a maximum dimension which is controllable through an option

Thanks! Was going to suggest this:
Is there a way to estimate the image size and add a warning after checking the arguments? Even if it causes a segfault, a warning will make it easier to debug (e.g., with options(warn = 2)).

[thomasp85]

This is the error given:

> ragg::agg_png(res = 200, units = "in")
Error: One or both dimensions exceed the maximum (50000px).
- Use options(ragg.max_dim = ...) to change the max
  Warning: May cause the R session to crash

[thomasp85]

Are you suggesting anything in addition?

[hsbadr]

Are you suggesting anything in addition?

I'll get back to you after debugging the current version of ragg on my system. I'll open a new issue, if I found the culprit with reproducible examples on other platforms.

[thomasp85]

Are we discussing the issue raised in the ggplot2 repo here? In the above I was more alluring to your comment about estimating image size and whether the current setup addresses that

[hsbadr]

Yes, what I was suggesting is to estimate the required memory size and return a warning if it exceeds a limit; though, the new ragg.max_dim option may do it.

[hsbadr]

The good thing about memory check is that you may use the available/physical memory as a dynamic limit.

[thomasp85]

The default max is set so high that if you ever reach it you are in "here-be-dragons" territory and on your own...

I don't know of any way to reliably estimate memory available to a process so I'd rather not go that route

[hsbadr]

I don't know of any way to reliably estimate memory available to a process so I'd rather not go that route

The estimated memory size depends on the data type and dimensions. It's an easy estimate (e.g., multiply the type size, e.g. 8 bytes, by the total number of pixels) and can be compared to a fraction of the available memory (e.g., a memory limit of 50% or so).

[hsbadr]

There's also a memuse package, but I haven't tested/used it.