copy lists that are known to be finite (though not necessarily
proper).
* debug.c (scm_procedure_source): Don't have scm_unmemocopy treat
a closure's argument list like an expression of a body.
* eval.c (unmemoize_expression, unmemoize_exprs, unmemoize_and,
unmemoize_begin, unmemoize_case, unmemoize_cond, unmemoize_delay,
unmemoize_do, unmemoize_if, unmemoize_lambda, unmemoize_let,
unmemoize_letrec, unmemoize_letstar, unmemoize_or,
unmemoize_set_x, unmemoize_apply, unmemoize_atcall_cc,
unmemoize_at_call_with_values, unmemoize_future, sym_atslot_ref,
unmemoize_atslot_ref, sym_atslot_set_x, unmemoize_atslot_set_x,
unmemoize_builtin_macro): New static functions and symbols.
(scm_unmemocopy): Rewritten in terms of the above. scm_unmemocopy
now has a slightly different meaning: The memoized form that is
receives as its argument is now interpreted as a sequence of
expressions from a body.
(unmemocar, scm_unmemocar): Since the whole functionality of
unmemocar and scm_unmemocar is not needed any more, scm_unmemocar
has its old content back and is deprecated, while unmemocar has
been removed.
(SCM_BIT7): Removed.
(CEVAL): For unmemoizing a single expression, call
unmemoize_expression instead of scm_unmemocopy, which now expects
a sequence of body expressions. Eliminated unnecessary empty
environment frame when executing let* forms. Eliminated
unmemoization step from evaluator.
macroexp and made static. Added new version of scm_macroexp that
emits a deprecation warning and then calls macroexp.
(scm_m_undefine): Issue deprecation warning.
misplaced syntactic keywords. This will not work unless guile's
defmacro feature is deprecated.
(scm_m_case): Fixed a bug that caused the list of labels to grow
with every case form.
(scm_m_quote, scm_m_atslot_ref, SCM_CEVAL): Changed the byte code
representation of 'quote' and '@slot-ref' to an improper list.
This reduces execution time, the number of cells used to hold the
memoized code, and thus also reduces garbage collection time.
(scm_unmemocopy): Use unmemoize_quote for quote expressions.
(SCM_CEVAL): Changed macro handling to also work with macros that
return improper lists. Added an assertion, that the code returned
by a macro transformer will not lead to cycles in the memoized
code.
file.
* eval.c (scm_ilookup, scm_unbound_variable_key,
error_unbound_variable, scm_lookupcar1, scm_lookupcar): Moved to
the definitions used for execution, since that's where they will
belong to later.
(scm_m_define): Change order to first create binding and
evaluating the expression afterwards.
(scm_m_set_x): Memoize complete set! expression. Only leave
symbols if no binding exists at memoization time. Throw error if
assigning to a syntactic keyword.
(lazy_memoize_variable): New function.
(CEVAL): When execution set!, perform lazy memoization if
unmemoized symbol is detected.
* libguile/modules.c (module_variable): Return variables with
unbound value.
* libguile/tags.h: Fix comment.
* test-suite/tests/syntax.test: Add test case to check the correct
handling of define expressions.
* deprecated.h (SCM_IFRINC, SCM_ICDR, SCM_IFRAME, SCM_IDIST,
SCM_ICDRP), eval.c (SCM_IFRINC, SCM_ICDR, SCM_IFRAME, SCM_IDIST,
SCM_ICDRP), eval.h (SCM_ICDR, SCM_IFRINC, SCM_IFRAME, SCM_IDIST,
SCM_ICDRP): Deprecated and added to deprecated.h. Moved from
eval.h to eval.c.
* deprecated.c (scm_isymnames), deprecated.h (scm_isymnames,
SCM_ISYMNUM, SCM_ISYMCHARS), eval.c (SCM_ISYMNUM, isymnames,
scm_unmemocopy, CEVAL), print.c (scm_isymnames), tags.h
(SCM_ISYMNUM, scm_isymnames, SCM_ISYMCHARS): Deprecated
scm_isymnames, SCM_ISYMNUM and SCM_ISYMCHARS and added to
deprecated.[hc]. Moved scm_isymnames from print.c to eval.c and
renamed to isymnames. Moved SCM_ISYMNUM from tags.h to eval.c and
renamed to ISYMNUM.
* eval.c (scm_i_print_iloc, scm_i_print_isym), eval.h
(scm_i_print_iloc, scm_i_print_isym), print.c (scm_iprin1):
Extracted printing of ilocs and isyms to guile internal functions
scm_i_print_iloc, scm_i_print_isym of eval.c.
instructions when dispatching - just always dispatch on the
instruction code, which is common for short and long instructions.
Further, removed unnecessary goto statements and added comment.
long instructions when dispatching - just always dispatch on the
instruction code, which is common for short and long instructions.
Further, removed unnecessary goto statements, fixed indentation
and replaced SCM_IMP predicates by SCM_NULLP.
SCM_DEBUGGINGP:
* debug.h (scm_debug_mode_p, scm_debug_mode, SCM_DEBUGGINGP),
eval.c (scm_debug_mode_p): Deprecated scm_debug_mode and
SCM_DEBUGGINGP. Provided scm_debug_mode_p instead, to have one
single interface that also matches the naming conventions.
Probably scm_debug_mode_p should be part of the private interface
anyway.
* debug.h (scm_debug_mode_p), backtrace.c (display_error_body),
eval.c (SCM_APPLY, scm_trampoline_0, scm_trampoline_1,
scm_trampoline_2): Change uses of scm_debug_mode or SCM_DEBUGGINGP
to scm_debug_mode_p.
Deprecate direct access to scm_ceval, scm_deval and scm_ceval_ptr:
* eval.h (scm_ceval, scm_deval, scm_ceval_ptr), debug.h
(scm_ceval_ptr): Deprecated. Moved declaration of scm_ceval_ptr
from debug.h to eval.h.
* debug.h (SCM_RESET_DEBUG_MODE): Don't access scm_ceval_ptr any
more, just leave it with setting scm_debug_mode_p, which is
equivalent for practical purposes.
* deprecated.h (SCM_XEVAL, SCM_XEVALCAR): Call scm_i_eval_x
instead of *scm_ceval_ptr. Leave all evaluating to scm_i_eval_x.
* gdbint.c (gdb_eval): Call scm_i_eval_x instead of scm_ceval.
* eval.c (ceval, deval, scm_ceval, scm_deval): Made scm_ceval
static and renamed it to ceval throughout. Provide a new exported
but deprecated function scm_ceval as a wrapper for backwards
compatibility. The same is done for the deval/scm_deval pair of
functions.
* eval.c (CEVAL, SCM_CEVAL): Renamed SCM_CEVAL to CEVAL
throughout. Defined CEVAL to ceval or deval, based on compilation
phase.
* eval.c (SCM_XEVAL, SCM_XEVALCAR): Dispatch on scm_debug_mode_p
to ceval and deval instead of calling *scm_ceval_ptr.
* eval.c (dispatching_eval): New deprecated static function.
* eval.c (scm_ceval_ptr): Initialized to dispatching_eval in order
to emulate its old behaviour as closely as possible.
Change the evaluator such that only expressions for which pair? is
true are passed to CEVAL, and such that all other expressions are
evaluated outside of CEVAL:
* eval.c (EVAL): New, provided in analogy to EVALCAR. Evaluate an
expression that is assumed to be memoized already. All but
expressions of the form '(<form> <form> ...)' are evaluated inline
without calling an evaluator.
* eval.c (SCM_XEVAL, SCM_XEVALCAR, EVALCAR): Evaluate all but
expressions of the form '(<form> <form> ...)' inline without
calling an evaluator.
* eval.c (scm_i_eval_x, scm_i_eval, scm_ceval, scm_deval): Handle
the special case of unmemoized symbols passed on the top level.
* eval.c (CEVAL): Change calls to CEVAL to EVAL, except where it
is known that the expression passed to CEVAL is of the form
'(<form> <form> ...)'. Remove handling of the tc7-objects, since
now it is known that the input expression of CEVAL is a pair.
from the memoization part of the file.
(copy_tree): New static function
(scm_copy_tree): Rewritten to fix two kinds or bugs: First, cyclic
structures are detected now and will lead to an exception instead
of forcing guile to run in an endless loop, using up all the
system's memory. Second, arrays in the cdr of an improper list
are now copied. See the new test cases in eval.test.
* test-suite/tests/eval.test: Added tests which reflect the recent
fixes to copy-tree.
FORMS) also when a letrec is constructed, not only when no definitions
are found. Do not return rewritten expression to emphasize the
in-place rewriting. Changed all users.
(m_body): Fixed comment.
(scm_m_define): Don't generate memoized code for definitions that
are not on the top level. As a consequence, no memoized code at
all is generated for definitions any more: Top level definitions
are executed immediately during memoization and internal
definitions are handled separately in m_expand_body.
(scm_unmemocopy, unmemocopy): Removed code for unmemoizing
definitions. Consequently, there is no unmemoizing code any more
that might modify the environment. Thus, the old scm_unmemocopy
is removed and the old unmemocopy is renamed to scm_unmemocopy.
(SCM_CEVAL): The SCM_IM_DEFINE keyword can no longer occur in
memoized code. Call EVALCAR for continuations. Prefer !SCM_NULLP
over SCM_NIMP in places, where the argument is known to be part of
a proper list.
(lookup_global_symbol, lookup_symbol): New static functions.
(s_test, s_bindings, s_duplicate_bindings, s_variable): Removed.
(try_macro_lookup, literal_p): Use lookup_symbol instead of
creating a temporary pair for scm_lookupcar.
(scm_unmemocar, unmemocar): Renamed scm_unmemocar to unmemocar,
created deprecated wrapper function scm_unmemocar.
(SCM_VALIDATE_NON_EMPTY_COMBINATION, scm_sym_else,
scm_sym_unquote, scm_sym_uq_splicing, scm_sym_enter_frame,
scm_sym_apply_frame, scm_sym_exit_frame, scm_sym_trace, f_apply,
undefineds, sym_three_question_marks): Moved around without
modifications.
* eval.c, eval.h (scm_macroexp, scm_unmemocar): Deprecated.
(canonicalize_define, scm_m_define): The check for a bad
expression is performed in canonicalize_define now.
(try_macro_lookup, expand_user_macros, is_system_macro_p): New
static helper functions for m_expand_body.
(m_expand_body): Use ASSERT_SYNTAX to signal syntax errors. Only
expand user defined macros. Fixed handling of the definition/
expression boundary. Fixed handling of definitions grouped with
'begin. Use canonicalize_define to expand definitions.
* test-suite/tests/syntax.test: Tests that check for the correct
handling of internal defines with begin work now.
public use of scm_m_expand_body in eval.h. In eval.c, renamed
scm_m_expand_body to m_expand_body and made it static. Added
deprecated wrapper scm_m_expand_body.
(scm_eval_body, SCM_CEVAL, SCM_APPLY): Use m_expand_body instead
of scm_m_expand_body.
(s_body): Removed.
(scm_m_expand_body): Fixed core dump when passing a body with
defines, but without expressions (see additions to syntax.test).
Use ASSERT_SYNTAX to signal syntax errors.
* test-suite/tests/syntax.test (exception:missing-body-expr): New.
Renamed section 'define' to 'top-level define' and added a new
section 'internal define' with some tests.
handled in scm_m_body any more, but rather in scm_m_lambda.
(scm_m_body, memoize_named_let, scm_m_let, scm_m_letstar,
scm_m_letrec, scm_m_expand_body): Check for validity is done by
calling functions of scm_m_body.
(scm_m_lambda): Avoid unnecessary consing when creating the
memoized code.
(s_empty_combination, error_unbound_variable): New static
identifiers.
(SCM_VALIDATE_NON_EMPTY_COMBINATION, SCM_EVALIM2, scm_lookupcar1):
Use ASSERT_SYNTAX, syntax_error or error_unbound_variable to
signal syntax errors.
(SCM_CEVAL): Separated handling of evaluator bytecodes and other
scheme objects.
* test-suite/tests/syntax.test (exception:missing/extra-expr-misc):
Removed.
(exception:illegal-empty-combination): New.
(exception:missing/extra-expr): Unified capitalization.
Adapted test for '()' to the new way of error reporting.
Grouped together with unmemocopy, without modifications.
(build_binding_list, unmemocopy): Renamed names of list arguments
and variables to reflect the actual order of the list elements.
(scm_m_nil_cond, scm_m_atfop, scm_m_undefine): Add comments. Use
ASSERT_SYNTAX to signal syntax errors. Avoid unnecessary consing
when creating the memoized code.
scm_m_generalized_set_x): Use ASSERT_SYNTAX to signal syntax
errors. Avoid unnecessary consing when creating the memoized
code.
(scm_m_generalized_set_x): Let scm_m_set_x handle the R5RS
standard case. Make sure line and file information are copied to
every created expression.
* test-suite/tests/syntax.test (exception:bad-var): Removed.
Adapted tests for 'set!' to the new way of error reporting.
ASSERT_SYNTAX to signal syntax errors. Avoid unnecessary consing
when creating the memoized code.
(scm_m_atbind): Reversed the order, in which the init expressions
are stored and executed. The order of execution is now equal to
the order in which the initializers of the let-forms are executed.
Use check_bindings and transform_bindings.
(SCM_CEVAL): Eliminated SCM_NIMP in favor of more appropriate
!SCM_NULLP. Added some comments.
* test-suite/tests/dynamic-scope.test (exception:missing-expr):
Introduced temporarily until all memoizers use the new way of error
reporting.
(scm_m_begin): Added comment.
(scm_m_or): Use ASSERT_SYNTAX to signal syntax errors. Avoid
unnecessary consing when creating the memoized code.
(iqq, scm_m_quasiquote, scm_m_quote): Use ASSERT_SYNTAX to signal
syntax errors. Be more specific about the kind of error that was
detected.
(scm_m_quote, unmemocopy): As an optimization, vector constants
are now inserted unquoted into the memoized code. During
unmemoization the quotes are added again to provide syntactically
correct code.
* test-suite/tests/syntax.test (exception:missing/extra-expr,
exception:missing/extra-expr-misc): Renamed
exception:missing/extra-expr to exception:missing/extra-expr-misc.
(exception:missing/extra-expr-syntax,
exception:missing/extra-expr): Renamed
exception:missing/extra-expr-syntax to
exception:missing/extra-expr.
scm_m_expand_body, check_bindings): Extracted syntax checking of
bindings to new static function check_bindings.
(scm_m_let, memoize_named_let): Extracted handling of named let to
new static function memoize_named_let.
(transform_bindings, scm_m_let, scm_m_letstar, scm_m_letrec): Use
ASSERT_SYNTAX to signal syntax errors. Be more specific about the
kind of error that was detected. Avoid use of SCM_CDRLOC. Avoid
unnecessary consing when creating the memoized code.
* test-suite/lib.scm (exception:bad-variable): New.
* test-suite/tests/syntax.test (exception:bad-binding,
exception:duplicate-binding): New.
(exception:duplicate-bindings): Removed.
Adapted tests for 'let', 'let*' and 'letrec' to the new way of
error reporting.
New static identifiers.
(s_clauses, s_formals, s_duplicate_formals): Removed.
(scm_m_lambda): Use ASSERT_SYNTAX to signal syntax errors. Be more
specific about the kind of error that was detected. Prepare for
easier integration of changes for separated memoization.
* test-suite/tests/syntax.test (define exception:bad-formal, define
exception:duplicate-formal): New.
(exception:duplicate-formals): Removed.
(exception:bad-formals): Adapted to the new way of error
reporting.
Adapted tests for 'lambda' to the new way of error reporting.
(scm_m_case): Call scm_c_memq instead of implementing it inline.
(scm_m_define): Added comment about how we check for duplicate
formals.
(scm_m_do): Added check for duplicate bindings.
(scm_m_if): Use ASSERT_SYNTAX to signal syntax errors. Avoid
unnecessary consing when creating the memoized code.
(scm_c_improper_memq, c_improper_memq, scm_m_lambda): Renamed
scm_c_improper_memq to c_improper_memq, since it is not exported.
(transform_bindings): Call scm_c_memq rather than
scm_c_improper_memq.
(SCM_CEVAL): Simplified handling of SCM_IM_IF forms.
* test-suite/tests/syntax.test (exception:missing/extra-expr-syntax):
New, introduced temporarily until all memoizers use the new way of
error reporting.
Adapted tests for 'if' to the new way of error reporting.
static identifiers.
(scm_m_do): Use ASSERT_SYNTAX to signal syntax errors. Be more
specific about the kind of error that was detected. Avoid use of
SCM_CDRLOC. Avoid unnecessary consing when creating the memoized
code, this way also making sure that file name, line number
information etc. remain available.
identifiers.
(scm_m_define): Use ASSERT_SYNTAX to signal syntax errors. Prefer
R5RS terminology for the naming of variables. Be more specific
about the kind of error that was detected. Make sure file name,
line number etc. are added to all freshly created expressions.
Avoid unnecessary consing when creating the memoized code.
* test-suite/tests/syntax.test (exception:missing-expr,
exception:extra-expr): New.
Adapted tests for 'begin' to the new way of error
reporting.
s_bad_cond_clause, s_missing_recipient): New static identifiers.
(s_extra_case_clause): Removed.
(scm_m_case, scm_m_cond): If a clause appears after an else
clause, report a misplaced else clause.
(scm_m_cond): Use ASSERT_SYNTAX to signal syntax errors. Be more
specific about the kind of error that was detected. Handle bound
'else and '=>. Avoid unnecessary consing when creating the
memoized code.
(scm_m_cond, unmemocopy, SCM_CEVAL): Use SCM_IM_ELSE to memoize
the syntactic keyword 'else and SCM_IM_ARROW to memoize the
syntactic keyword '=>.
* test-suite/tests/syntax.test (exception:misplaced-else-clause,
exception:bad-cond-clause): New.
(exception:bad/missing-clauses, exception:extra-case-clause):
Removed.
Adapted tests for 'case' and 'cond' to the new way of error
reporting.
The tests that check if cond is hygienic pass now.
* libguile/print.c (scm_isymnames): Add names for the new memoizer
codes.
* libguile/eval.c (s_missing_clauses, s_bad_case_clause,
s_extra_case_clause, s_bad_case_labels, s_duplicate_case_label,
literal_p): New static identifiers.
(scm_m_case): Use ASSERT_SYNTAX to signal syntax errors. Be more
specific about the kind of error that was detected. Check for
duplicate case labels. Handle bound 'else. Avoid unnecessary
consing when creating the memoized code.
(scm_m_case, unmemocopy, SCM_CEVAL): Use SCM_IM_ELSE to memoize
the syntactic keyword 'else.
* test-suite/tests/syntax.test (exception:bad-expression,
exception:missing-clauses, exception:bad-case-clause,
exception:extra-case-clause, exception:bad-case-labels): New.
Added some tests and adapted tests for 'case' to the new way of
error reporting.
