Add interactive visualization pages for EncodeUtf8, DecodeUtf8, AppendString, and EqualsString with benchmark data plotting and model fit overlays. Update portal index with string builtin links.
All Text benchmarks now use 4-byte-only characters (worst case for costing). Since ceil(bytes/4) = chars for 4-byte characters, the same fitted parameters work for both char-count (A/B/C) and byte-length (D/E) sizing. This eliminates the separate 4Byte benchmark functions, reducing the suite from 7 to 4 benchmarks. Add named-with-wrapper benchmark helpers in Common.hs and a 4-byte text generator in Generators.hs.
Change ExMemoryUsage for Text from character count to byte-length-based measurement (ceil(bytes/8) words), consistent with ByteString. Preserve memoryUsage 0 for empty Text as a special case. Update encodeUtf8 and decodeUtf8 memory model slopes to reflect byte-based sizing. Add ExMemoryUsage property tests for Text values.
Gate string builtin costing by semantics variant via TextCostingByteLength: A/B/C use character count, D/E use ceil(byte-length/4) half-words. The byte-length metric is O(1) to compute (vs O(n) char traversal) and more accurately reflects actual work done by string operations. Add D/E cost model files (builtinCostModelD.json, builtinCostModelE.json). Update test golden files for the new variant D type signatures and conformance budget expectations.
`plutus-metatheory` only builds with `ghc-9.6` (currently) so anything that depends on it should also not be buildable.
`plutus-metatheory` only builds with `ghc-9.6` (currently) so anything that depends on it should also not be buildable.
Add benchOneTextString4Byte for EncodeUtf8: generates strings using only 4-byte UTF-8 characters (U+10000..U+10FFFF) with char-count x-axis. For these inputs, ceil(bytes/4) = char_count exactly, so fitting with denominator 4 should produce parameters matching the char-count model. Adjust size ranges to match master's effective range (which was doubled by the old 2*n generator bug): oneArgumentSizes: [0,100..10000] -> [0,200..20000] (101 points) twoArgumentSizes: [0,250..5000] -> [0,500..10000] (21 points)
Document why VBuiltin terms don't need global shifting during discharge, and note the unchecked Word64 overflow in shiftNamedDeBruijn.
Replace the two-pass dischargeCekValue implementation (discharge + shiftTermBy post-pass) with a single-pass approach that threads a global shift parameter through goValue/goValEnv. This avoids a separate traversal for shifting and handles truly free variables (not found in the environment) consistently. - Add shiftNamedDeBruijn utility to PlutusCore.DeBruijn - Thread `global` shift parameter through goValue and goValEnv - Delete the standalone shiftTermBy function - Add 4 new tests for truly free vars past non-empty environments
VBuiltin values can be stored in an environment and later discharged under additional binders. The previous code returned VBuiltin terms without shifting, causing variable capture. Add shiftTerm helper to shift free variables in discharged VBuiltin terms, add bang pattern to shiftNamedDeBruijn for consistency, and add a VBuiltin test case reproducing basetunnel's example from issue #7526.
Add shift==idx boundary test verifying bound variable detection, and VConstr test verifying free variables in constructor arguments are shifted correctly under lambdas.
Document the variable capture bug fix in the changelog.
Add 8 test cases covering variable capture scenarios in dischargeCekValue: - Free variables under 1, 2, and 3 lambdas - Deeply indexed free variables - Multiple free variables in the same term - Nested environment structures Tests verify that free variables are correctly shifted to prevent capture when terms are discharged from the evaluation environment. Tests for #7526
Add shiftTermBy function to correctly shift free variables when discharging values from the environment. Previously, free variables in discharged terms could be captured by outer lambdas, causing incorrect variable references in the output term. The fix tracks binding depth during discharge and shifts free variables (those with indices beyond the current binding depth) by the appropriate amount to maintain correct scoping. Resolves #7526
