Color diffusion mode

[itamarst]

This does a few things:

1. Makes differential mode less saturated red and blue, so it's easier to read.
2. Adds a new color mode, tentatively called "color-diffusion."

The idea is that I want to visually draw the viewer's eyes towards the wider frames. Much like a differential flamegraph pulls you towards the places where things got worse (or better) with red (or blue), this highlights the bad bits in a normal flamegraph, i.e. the wider the frame the redder it gets.

So some questions for you:

1. Do you have a better name?
2. Are you willing to accept this? The alternative is some sort of plugable color generation API, but it would have to take a lot of inputs to cover all cases.

[jasonrhansen]

I personally think the color diffusion mode is a nice addition.

I also like the look of the less saturated differential flamegraphs. It makes the text more readable when the sample diffs are larger. On the other hand, I think it makes it harder to see the paler hues. If the number of samples only differs by a small amount, it almost looks like white. At least on my monitor, in the second image below I can barely see the lightest red and blue.

Before change

After change

[jonhoo]

It would be really good to see some more before/after pictures to judge the effect of this. In general I'm for the change though.

[codecov]

Codecov Report

Merging #165 into master will increase coverage by 1.33%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #165      +/-   ##
==========================================
+ Coverage    88.5%   89.84%   +1.33%     
==========================================
  Files          16       16              
  Lines        2228     2226       -2     
==========================================
+ Hits         1972     2000      +28     
+ Misses        256      226      -30

Impacted Files	Coverage Δ
src/flamegraph/color/mod.rs	82.97% <100%> (+9.92%)	⬆️
src/flamegraph/mod.rs	95.8% <100%> (+0.04%)	⬆️
src/collapse/perf.rs	97.39% <0%> (ø)	⬆️
src/flamegraph/color/palettes.rs	99.63% <0%> (+0.36%)	⬆️
src/flamegraph/svg.rs	97.88% <0%> (+0.7%)	⬆️
src/flamegraph/color/palette_map.rs	93.8% <0%> (+0.88%)	⬆️
src/flamegraph/attrs.rs	92.92% <0%> (+1.01%)	⬆️
src/collapse/sample.rs	92.85% <0%> (+1.02%)	⬆️
src/collapse/vtune.rs	91.66% <0%> (+1.38%)	⬆️
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2aeab32...090d694. Read the comment docs.

[itamarst]

Looks like I can get some more example data from https://github.com/corpaul/flamegraphdiff/tree/master/demos, let me see what they look like.

[itamarst]

Nope, looks like artificial dataset, or at least just one or two function tweaked.

[itamarst]

OK, here is a more realistic example, an optimization I just tried for my Python memory profiler (yes, I'm generating flamegraphs for my flamegraph-generating project 😁 ).

Now, you might say "where did all the blue go!" but if you mouse over the you'll see it's like 0.20% improvement and other similarly tiny changes. 0.20% improvement is mostly noise, so having it be less visible is fine. And maybe it should be slightly more blue.

comparison.zip

[jonhoo]

I think slightly more contrast might be good, but apart from that I'm happy with it :)

[jonhoo]

Beautiful, thank you!

[jonhoo]

Ah, could you also add an entry to the "Added" section under "Unreleased" in CHANGELOG.md briefly describing the change and with a link to the PR?