Incorrect savefig size when using size = ( , )

[michaelwang314]

using Plots
gr()

plot(size = (800, 800))
plot!([1, 2], [4, 7])

savefig("TEST.png")

When I run this code, the saved image comes out as the incorrect size 799x800. If instead I do, for example,

plot(size = (600, 600))

I get images of correct size 600x600. While this isn't a big issue, I found that this has some problems with certain codecs when converting a sequence of images, say of size 799x800, to a movie (they require images with even dimensions).

EDIT: If I save as (500, 500), I get a 499x499 image. If I then do (501, 501), I still get 499x499. If I then do (502, 502), I get a 501x501 image. Finally, if I go back to (501, 501), it correctly saves a 501x501 image.

[daschw]

Thanks for reporting, this is really weird.
@jheinen I think this is an issue on the Plots side, and not with GR, right?

[michaelwang314]

Perhaps not surprisingly, the CirclePlot example on the Animations page with the addition of size --> (800, 800) generates a gif of incorrect size (799, 800). And similarly size --> (500, 500) gives (499, 499), as above.

Just ran some tests with everything updated. I think the last step in my previous tests that gave the correct size might have been an error on my part, though I'm not sure.

using Plots
# saves an image of size (499, 499)
plt = plot(size = (500, 500))
savefig("test1.png")

# saves an image of size (500, 501)
plt = plot(size = (501, 501))
savefig("test2.png")

# saves an image of size (501, 501)
plt = plot(size = (502, 502))
savefig("test3.png")

# saves an image of size (500, 501)
plt = plot(size = (501, 501))
savefig("test4.png")

# saves an image of size (501, 500)
plt = plot(size = (501, 500))
savefig("test5.png")

The last two are a little strange (decreasing the height by 1px fixes the width). I get the same results on my laptop and desktop. Everything seems to be stored fine under the plot attributes plt.attr[:size]. Is the issue with how images/animations are saved (e.g. there's some weird rounding issue when the actual file is generated)?

If I systematically decrease one of the dimensions, say the height, I get

using Plots
cd(dirname(@__FILE__))
for i = 490 : 510
    plot(size = (500, i))
    savefig("test_$i.png")
end

input          output
(500, 490) --> (500, 489)
(500, 491) --> (500, 491) | Good!
(500, 492) --> (500, 492) | Good!
(500, 493) --> (500, 493) | Good!
(500, 494) --> (500, 494) | Good!
(500, 495) --> (500, 495) | Good!
(500, 496) --> (500, 495)
(500, 497) --> (500, 497) | Good!
(500, 498) --> (500, 498) | Good!
(500, 499) --> (500, 499) | Good!
(500, 500) --> (499, 499)
(500, 501) --> (499, 501)
(500, 502) --> (499, 501)
(500, 503) --> (500, 502)
(500, 504) --> (499, 503)
(500, 505) --> (500, 505) | Good!
(500, 506) --> (499, 505)
(500, 507) --> (500, 507) | Good!
(500, 508) --> (499, 508)
(500, 509) --> (500, 509) | Good!
(500, 510) --> (499, 509)

I don't see a pattern...

[BeastyBlacksmith]

Can you test that with a different backend?

[michaelwang314]

Can you test that with a different backend?

The above tests were with gr(). With plotly(), all the above tests work fine. With pyplot(), there are some issues but not as many as gr(). Below are the tests with pyplot():

input          output
(500, 490) --> (500, 490) | Good!
(500, 491) --> (500, 491) | Good!
(500, 492) --> (500, 492) | Good!
(500, 493) --> (500, 493) | Good!
(500, 494) --> (500, 494) | Good!
(500, 495) --> (500, 495) | Good!
(500, 496) --> (500, 496) | Good!
(500, 497) --> (500, 497) | Good!
(500, 498) --> (500, 498) | Good!
(500, 499) --> (500, 499) | Good!
(500, 500) --> (500, 500) | Good!
(500, 501) --> (500, 501) | Good!
(500, 502) --> (500, 501)
(500, 503) --> (500, 503) | Good!
(500, 504) --> (500, 504) | Good!
(500, 505) --> (500, 505) | Good!
(500, 506) --> (500, 505)
(500, 507) --> (500, 507) | Good!
(500, 508) --> (500, 508) | Good!
(500, 509) --> (500, 509) | Good!
(500, 510) --> (500, 509)

(800, 800) --> (800, 800) | Good!

[michaelwang314]

Oh I forgot one more test with pyplot(). This is the only one that fails:

(502, 502) --> (501, 501) 

[amp1098]

Just sharing that this bug is still present.