Embedded fonts different across runs

I’ve noticed that if I generate the same PDF twice in a row, the output is different because the fonts get embedded with some kind of random prefix. For example, if I keep the CreationDate and ModDate fixed by setting the metadata:

doc.info.CreationDate = new Date(Date.parse('2018-09-15'));
doc.info.ModDate = new Date(Date.parse('2018-09-15'));

Then if I produce the same document twice it will contain diffs like these:

--- /tmp/one.hex        2018-09-17 18:32:54.000000000 +0200
+++ /tmp/two.hex        2018-09-17 18:33:00.000000000 +0200
@@ -37,7 +37,7 @@
 00000240: 3e0a 656e 646f 626a 0a31 3320 3020 6f62  >.endobj.13 0 ob
 00000250: 6a0a 3c3c 0a2f 5479 7065 202f 466f 6e74  j.<<./Type /Font
 00000260: 4465 7363 7269 7074 6f72 0a2f 466f 6e74  Descriptor./Font
-00000270: 4e61 6d65 202f 4552 4841 564e 2b50 6c61  Name /ERHAVN+Pla
+00000270: 4e61 6d65 202f 444b 4a49 4c57 2b50 6c61  Name /DKJILW+Pla
 00000280: 7966 6169 7244 6973 706c 6179 2d52 6567  yfairDisplay-Reg
 00000290: 756c 6172 0a2f 466c 6167 7320 340a 2f46  ular./Flags 4./F
 000002a0: 6f6e 7442 426f 7820 5b2d 3636 3820 2d32  ontBBox [-668 -2
@@ -52,7 +52,7 @@
 00000330: 6a0a 3c3c 0a2f 5479 7065 202f 466f 6e74  j.<<./Type /Font
 00000340: 0a2f 5375 6274 7970 6520 2f43 4944 466f  ./Subtype /CIDFo
 00000350: 6e74 5479 7065 320a 2f42 6173 6546 6f6e  ntType2./BaseFon
-00000360: 7420 2f45 5248 4156 4e2b 506c 6179 6661  t /ERHAVN+Playfa
+00000360: 7420 2f44 4b4a 494c 572b 506c 6179 6661  t /DKJILW+Playfa
 00000370: 6972 4469 7370 6c61 792d 5265 6775 6c61  irDisplay-Regula
 00000380: 720a 2f43 4944 5379 7374 656d 496e 666f  r./CIDSystemInfo
 00000390: 203c 3c0a 2f52 6567 6973 7472 7920 2841   <<./Registry (A
@@ -66,8 +66,8 @@
 00000410: 3e3e 0a65 6e64 6f62 6a0a 3620 3020 6f62  >>.endobj.6 0 ob
 00000420: 6a0a 3c3c 0a2f 5479 7065 202f 466f 6e74  j.<<./Type /Font
 00000430: 0a2f 5375 6274 7970 6520 2f54 7970 6530  ./Subtype /Type0
-00000440: 0a2f 4261 7365 466f 6e74 202f 4552 4841  ./BaseFont /ERHA
-00000450: 564e 2b50 6c61 7966 6169 7244 6973 706c  VN+PlayfairDispl
+00000440: 0a2f 4261 7365 466f 6e74 202f 444b 4a49  ./BaseFont /DKJI
+00000450: 4c57 2b50 6c61 7966 6169 7244 6973 706c  LW+PlayfairDispl
 00000460: 6179 2d52 6567 756c 6172 0a2f 456e 636f  ay-Regular./Enco
 00000470: 6469 6e67 202f 4964 656e 7469 7479 2d48  ding /Identity-H
 00000480: 0a2f 4465 7363 656e 6461 6e74 466f 6e74  ./DescendantFont
@@ -76,7 +76,7 @@
 000004b0: 656e 646f 626a 0a31 3720 3020 6f62 6a0a  endobj.17 0 obj.
 000004c0: 3c3c 0a2f 5479 7065 202f 466f 6e74 4465  <<./Type /FontDe
 000004d0: 7363 7269 7074 6f72 0a2f 466f 6e74 4e61  scriptor./FontNa
-000004e0: 6d65 202f 4e52 474a 5454 2b51 7561 7474  me /NRGJTT+Quatt
+000004e0: 6d65 202f 4c4f 4554 5a54 2b51 7561 7474  me /LOETZT+Quatt
 000004f0: 726f 6365 6e74 6f0a 2f46 6c61 6773 2034  rocento./Flags 4
 00000500: 0a2f 466f 6e74 4242 6f78 205b 2d33 3020  ./FontBBox [-30
 00000510: 2d32 3630 2031 3131 3620 3834 395d 0a2f  -260 1116 849]./
@@ -90,7 +90,7 @@
 00000590: 6a0a 3c3c 0a2f 5479 7065 202f 466f 6e74  j.<<./Type /Font
 000005a0: 0a2f 5375 6274 7970 6520 2f43 4944 466f  ./Subtype /CIDFo
 000005b0: 6e74 5479 7065 320a 2f42 6173 6546 6f6e  ntType2./BaseFon
-000005c0: 7420 2f4e 5247 4a54 542b 5175 6174 7472  t /NRGJTT+Quattr
+000005c0: 7420 2f4c 4f45 545a 542b 5175 6174 7472  t /LOETZT+Quattr
 000005d0: 6f63 656e 746f 0a2f 4349 4453 7973 7465  ocento./CIDSyste
 000005e0: 6d49 6e66 6f20 3c3c 0a2f 5265 6769 7374  mInfo <<./Regist
 000005f0: 7279 2028 4164 6f62 6529 0a2f 4f72 6465  ry (Adobe)./Orde
@@ -116,8 +116,8 @@
 00000730: 0a3e 3e0a 656e 646f 626a 0a39 2030 206f  .>>.endobj.9 0 o
 00000740: 626a 0a3c 3c0a 2f54 7970 6520 2f46 6f6e  bj.<<./Type /Fon
 00000750: 740a 2f53 7562 7479 7065 202f 5479 7065  t./Subtype /Type
-00000760: 300a 2f42 6173 6546 6f6e 7420 2f4e 5247  0./BaseFont /NRG
-00000770: 4a54 542b 5175 6174 7472 6f63 656e 746f  JTT+Quattrocento
+00000760: 300a 2f42 6173 6546 6f6e 7420 2f4c 4f45  0./BaseFont /LOE
+00000770: 545a 542b 5175 6174 7472 6f63 656e 746f  TZT+Quattrocento
 00000780: 0a2f 456e 636f 6469 6e67 202f 4964 656e  ./Encoding /Iden
 00000790: 7469 7479 2d48 0a2f 4465 7363 656e 6461  tity-H./Descenda
 000007a0: 6e74 466f 6e74 7320 5b31 3820 3020 525d  ntFonts [18 0 R]
@@ -125,7 +125,7 @@
 000007c0: 2052 0a3e 3e0a 656e 646f 626a 0a32 3120   R.>>.endobj.21
 000007d0: 3020 6f62 6a0a 3c3c 0a2f 5479 7065 202f  0 obj.<<./Type /
 000007e0: 466f 6e74 4465 7363 7269 7074 6f72 0a2f  FontDescriptor./
-000007f0: 466f 6e74 4e61 6d65 202f 4355 4845 5252  FontName /CUHERR
+000007f0: 466f 6e74 4e61 6d65 202f 4f59 4441 4e43  FontName /OYDANC
 00000800: 2b50 6c61 7966 6169 7244 6973 706c 6179  +PlayfairDisplay
 00000810: 2d42 6f6c 640a 2f46 6c61 6773 2034 0a2f  -Bold./Flags 4./
 00000820: 466f 6e74 4242 6f78 205b 2d37 3034 202d  FontBBox [-704 -
@@ -140,7 +140,7 @@
 000008b0: 626a 0a3c 3c0a 2f54 7970 6520 2f46 6f6e  bj.<<./Type /Fon
 000008c0: 740a 2f53 7562 7479 7065 202f 4349 4446  t./Subtype /CIDF
 000008d0: 6f6e 7454 7970 6532 0a2f 4261 7365 466f  ontType2./BaseFo
-000008e0: 6e74 202f 4355 4845 5252 2b50 6c61 7966  nt /CUHERR+Playf
+000008e0: 6e74 202f 4f59 4441 4e43 2b50 6c61 7966  nt /OYDANC+Playf
 000008f0: 6169 7244 6973 706c 6179 2d42 6f6c 640a  airDisplay-Bold.
 00000900: 2f43 4944 5379 7374 656d 496e 666f 203c  /CIDSystemInfo <
 00000910: 3c0a 2f52 6567 6973 7472 7920 2841 646f  <./Registry (Ado
@@ -165,7 +165,7 @@
 00000a40: 3130 2030 206f 626a 0a3c 3c0a 2f54 7970  10 0 obj.<<./Typ
 00000a50: 6520 2f46 6f6e 740a 2f53 7562 7479 7065  e /Font./Subtype
 00000a60: 202f 5479 7065 300a 2f42 6173 6546 6f6e   /Type0./BaseFon
-00000a70: 7420 2f43 5548 4552 522b 506c 6179 6661  t /CUHERR+Playfa
+00000a70: 7420 2f4f 5944 414e 432b 506c 6179 6661  t /OYDANC+Playfa
 00000a80: 6972 4469 7370 6c61 792d 426f 6c64 0a2f  irDisplay-Bold./
 00000a90: 456e 636f 6469 6e67 202f 4964 656e 7469  Encoding /Identi
 00000aa0: 7479 2d48 0a2f 4465 7363 656e 6461 6e74  ty-H./Descendant

(These diffs produced by dumping the hex with xxd one.pdf > one.hex and then comparing with diff -U 3 {one,two}.hex.)

It would be nice if running the same generation process twice in a row with the same inputs produced the exact same output (byte for byte). I’m committing these build artifacts to a gh-pages branch, and the churn in the PDFs is making the commits noisier than they need to be.

[blikblum]

The ‘xx’ was just a placeholder for actual date

[nrkn]

I think the best solution would be to generate a string from a hash of the font data. That way it should be unique to the font and also consistent across runs.