ponder-lab/ML: 0.52.103

Abstract

What's Changed Record three generator invariants established by wala/ML#907, #916 and #928 by @khatchad in https://github.com/ponder-lab/ML/pull/859 Give the allocation-site extractor a declining variant and move the callers that already expect a null onto it (wala/ML#925) by @khatchad in https://github.com/ponder-lab/ML/pull/860 Record that removing a summary body's return is a contract change by @khatchad in https://github.com/ponder-lab/ML/pull/861 Divert the rank predicate's shape read through the memo layer so a rank assert on an unpacked op result no longer overflows the stack (wala/ML#923) by @khatchad in https://github.com/ponder-lab/ML/pull/863 Say which direction a generator's legacy and record reads pair in by @khatchad in https://github.com/ponder-lab/ML/pull/864 Delete the producer delegation's definition-variable branch, which no allocation can reach (wala/ML#930) by @khatchad in https://github.com/ponder-lab/ML/pull/865 Resolve a class base written through a wildcard-bound name, so the class inherits the summary shell (wala/ML#938) by @khatchad in https://github.com/ponder-lab/ML/pull/866 Type tf.shape's result as a rank-1 int32 vector of the operand's rank (wala/ML#943) by @khatchad in https://github.com/ponder-lab/ML/pull/867 Resolve a base class by its module's scope and translate modules after the modules their bases come from (wala/ML#944, wala/ML#946, wala/ML#945) by @khatchad in https://github.com/ponder-lab/ML/pull/868 Declare a TRANSFORM type feed on Reshape, so a reshape of a dataflow-typed input keeps no unknown-dtype seed member beside the delivered dtype (wala/ML#940). by @khatchad in https://github.com/ponder-lab/ML/pull/872 Declare a pass-through type feed on ConvertToTensor, so a dataflow-typed value keeps no wholly unknown seed member through tf.convert_to_tensor and the summaries routed through it (wala/ML#947) by @khatchad in https://github.com/ponder-lab/ML/pull/873 Model tf.unstack in the tf.split form and declare a dtype feed on Stack, so a decoder's past keeps its dtype around the sampling loop (wala/ML#948, wala/ML#950). by @khatchad in https://github.com/ponder-lab/ML/pull/874 Declare a rule-carrying type feed on SliceBuiltinOperation, so a subscript of a dataflow-typed receiver takes the subscript's shape instead of the receiver's type or a wholly unknown seed (wala/ML#953) by @khatchad in https://github.com/ponder-lab/ML/pull/875 Model tf.keras.losses.SparseCategoricalCrossentropy, so a loss and everything computed from it take the predictions' dtype and the reduction's shape instead of reading ⊤ (wala/ML#951). by @khatchad in https://github.com/ponder-lab/ML/pull/876 Model np.cumsum, so a running sum reads its input's promoted dtype and the flattened or axis-kept shape instead of nothing (wala/ML#954). by @khatchad in https://github.com/ponder-lab/ML/pull/877 Empty a fed subscript result's pin, so a subscript of a dataflow-typed receiver keeps the feed's composed member alone rather than beside its seed's unresolved-dtype twin (wala/ML#957) by @khatchad in https://github.com/ponder-lab/ML/pull/878 Widen an elementwise operand's dtype when its points-to evidence misses a return arm of the call that defines it, so a seed no longer pairs one arm's dtype with another arm's shape (wala/ML#958) by @khatchad in https://github.com/ponder-lab/ML/pull/879 Model list repetition and concatenation in the pointer analysis, so an iterated [None] * n or xs + ys carries its operands' elements instead of no evidence at all (wala/ML#960) by @khatchad in https://github.com/ponder-lab/ML/pull/880 Read a tensor input whose points-to set is exactly the None constant as no tensor, so a runtime-dead if x is not None arm no longer leaks an unknown-dtype twin (wala/ML#961) by @khatchad in https://github.com/ponder-lab/ML/pull/881 Read a concat element produced by an operation that cannot execute on its None-only input as no tensor, at any position and on both axes (wala/ML#962). by @khatchad in https://github.com/ponder-lab/ML/pull/882 Bind a destructuring comprehension target, and give comprehensions and numpy summaries their caller's context (wala/ML#955) by @khatchad in https://github.com/ponder-lab/ML/pull/883 Model tf.scatter_nd, so a scatter takes its dtype from updates and its shape from shape (wala/ML#966). by @khatchad in https://github.com/ponder-lab/ML/pull/884 Make no field key for the None constant, so an attribute write whose receiver may be None no longer leaks as an element of every container that may be None (wala/ML#964). by @khatchad in https://github.com/ponder-lab/ML/pull/885 Feed a pass-through-family op's input when it is passed by keyword, not only by position (wala/ML#967). by @khatchad in https://github.com/ponder-lab/ML/pull/887 Model text reads, so a top-level open reaches the builtin and a dataset sliced from a file's lines gives its map callback a scalar string (wala/ML#969). by @khatchad in https://github.com/ponder-lab/ML/pull/888 Model tf.reverse_sequence, so its result keeps its input's shape and dtype instead of carrying no tensor type (wala/ML#965). by @khatchad in https://github.com/ponder-lab/ML/pull/886 Decide a φ arm that reaches its merge through a chain of blocks, so a folded guard drops its dead arm however many calls that arm makes (wala/ML#970). by @khatchad in https://github.com/ponder-lab/ML/pull/889 Bind a @click.option default under a constant key of its own class, so the comparison fold declines a guard over it while a shape read keeps its value (wala/ML#971). by @khatchad in https://github.com/ponder-lab/ML/pull/891 Suppress a dead call site's argument and result edges, so a helper dispatched on a string argument no longer receives the other arm's inputs (wala/ML#968). by @khatchad in https://github.com/ponder-lab/ML/pull/890 Model the generic tf.keras.layers.RNN over a user cell, so its output takes the cell's dtype with the time axis reinserted (wala/ML#973). by @khatchad in https://github.com/ponder-lab/ML/pull/893 Rewrap the shared-arm comment in the φ suppression loop (wala/ML#970). by @khatchad in https://github.com/ponder-lab/ML/pull/892 Full Changelog: https://github.com/ponder-lab/ML/compare/0.52.102...0.52.103