Modify: xmlReadFile() at IccProfileXml.cpp:877 & IccMpeXml.cpp:2400

[xsscx]

Maintainer Summary

2026-03-27 00:03:32 UTC

iccDEV v2.3.1.5+85663e4 (ASAN+UBSAN)

Profile: v5.00 NamedColorClass (nmcl)
Spectral PCS: 31-step biSpectral reflectance (400-700nm, BiSpectral 300-700nm, 41 steps)
Profile ID: 681d53eca55a99d11051f605e7a105c0

Observations

• The XML to Binary to XML round-trip is NOT semantically reversible for calculator-heavy profiles
• Original numerical intent/precision is preserved in float16
• Binary encoding according to ICC.2-2023 specification
• Data loss (calculator ops, labels) is by design -- the binary format stores evaluated values, not source expressions
• The libxml2 10MB limit is a Bug
  • Error: "xmlSAX2Characters: huge text node"
    • IccProfileXml.cpp:877 xmlReadFile(szFilename, NULL, 0)
    • IccMpeXml.cpp:2400 xmlReadFile(file.c_str(), NULL, 0)
    • Neither passes XML_PARSE_HUGE (0x80000) flag
• Potential observer label loss bug (enum mapping)
  • iccToXml maps it back to "Unknown observer"
  • (x-bar, y-bar, z-bar) is preserved

Profile Conversion Results

Numbers: PRESERVED (float32 precision, max rel err <5e-08)
Headers: PRESERVED (with expected normalization)
Tags: PRESERVED (6 tags, structurally equivalent)
Text: PRESERVED (copyright, description identical)
Spectral: PRESERVED (observer + illuminant + white point)
Calc Logic: LOST (93,600 operations evaluated to float16 values)
Observer: LOST ("CIE 1931 standard..." returned "Unknown observer")
NaN/INF: LOST (111,489 leaving 79 survivors after evaluation)
Format: CHANGED (tab to space, 8 to 12 decimal, line wrapping)

TODO

• Change to xmlReadFile(szFilename, NULL, XML_PARSE_HUGE)
• Research potential observer label loss bug (enum mapping)

Originally posted by @xsscx in #524

• Missing XML_PARSE_NOENT and XML_PARSE_NONET options