Add notes for 2022-10-28

content/posts/2022-10.md

@@ -672,4 +672,89 @@ $ pngquant /tmp/10568-125167.pdf.png
 - Spent some time looking at the MediaBox / CropBox thing in DSpace's `ImageMagickThumbnailFilter.java`
   - We need to make sure to put `-define pdf:use-cropbox=true` before we specify the input file or else it will not have any effect
 
+## 2022-10-27
+
+- I found out that we can use [pdfcpu to remove the CropBox from a PDF](https://pdfcpu.io/boxes/boxes_remove.html#examples) for testing:
+
+```console
+$ pdfcpu box rem -- "crop" in.pdf out.pdf
+```
+
+- I filed [an issue on DSpace](https://github.com/DSpace/DSpace/issues/8549) for the ImageMagick `CropBox` problem
+  - I decided that this is a bug that should be fixed separately from the "improving thumbnail quality" issue
+  - I made [a pull request](https://github.com/DSpace/DSpace/pull/8550) to fix the `CropBox` issue
+- I did more work on my [improved-dspace-thumbnails](https://github.com/alanorth/improved-dspace-thumbnails/) microsite to complement the DSpace thumbnail pull requests
+  - I am updating it to recommend using the PDF cropbox and "supersampling" with a higher density than 72
+  - I measured execution time of ImageMagick with `time` and found that the higher-density mode takes about five times longer on average
+  - I measured the [maximum heap memory of ImageMagick with Valgrind and Massif](https://stackoverflow.com/a/131346):
+
+```console
+$ valgrind --tool=massif magick convert ...
+```
+
+- Then I checked the results for each set of default DSpace thumbnail runs and "improved" thumbnail runs using `ms_print` (hacky way to get the max heap, I know):
+
+```console
+$ for file in memory-dspace/massif.out.49*; do ms_print "$file" | grep -A1 "    MB" | tail -n1 | sed 's/\^.*//'; done
+15.87
+16.06
+21.26
+15.88
+20.01
+15.85
+20.06
+16.04
+15.87
+15.87
+20.02
+15.87
+15.86
+19.92
+10.89
+$ for file in memory-improved/massif.out.5*; do ms_print "$file" | grep -A1 "    MB" | tail -n1 | sed 's/\^.*//'; done
+245.3
+245.5
+298.6
+245.3
+306.8
+245.2
+306.9
+245.5
+245.2
+245.3
+306.8
+245.3
+244.9
+306.3
+165.6
+```
+
+- Ouch, this shows that it takes about *fifteen times* more memory to do the "4x" density of 288!
+  - It seems more reasonable to use a "2x" density of 144:
+
+```console
+$ for file in memory-improved-144/*; do ms_print "$file" | grep -A1 "    MB" | tail -n1 | sed 's/\^.*//'; done
+61.80
+62.00
+76.76
+61.82
+77.43
+61.77
+77.48
+61.98
+61.76
+61.81
+77.44
+61.81
+61.69
+77.16
+41.84
+```
+
+- There's a really cool visualizer called massif-visualizer, but it isn't easy to parse
+
+## 2022-10-28
+
+- I finalized the code for the ImageMagick density change and made a [pull request](https://github.com/DSpace/DSpace/pull/8553) against DSpace 7.x
+
 <!-- vim: set sw=2 ts=2: -->