diff options
Diffstat (limited to 'lib/debug.nix')
-rw-r--r-- | lib/debug.nix | 119 |
1 files changed, 119 insertions, 0 deletions
diff --git a/lib/debug.nix b/lib/debug.nix new file mode 100644 index 000000000000..d627bc861abb --- /dev/null +++ b/lib/debug.nix @@ -0,0 +1,119 @@ +let lib = import ./default.nix; + +inherit (builtins) trace attrNamesToStr isAttrs isFunction isList isInt + isString isBool head substring attrNames; + +inherit (lib) all id mapAttrsFlatten elem; + +in + +rec { + + + # Wrapper aroung the primop `addErrorContext', which shouldn't used + # directly. It evaluates and returns `val', but if an evaluation + # error occurs, the text in `msg' is added to the error context + # (stack trace) printed by Nix. + addErrorContext = + if builtins ? addErrorContext + then builtins.addErrorContext + else msg: val: val; + + addErrorContextToAttrs = lib.mapAttrs (a : v : lib.addErrorContext "while evaluating ${a}" v); + + + traceVal = if builtins ? trace then x: (builtins.trace x x) else x: x; + traceXMLVal = if builtins ? trace then x: (builtins.trace (builtins.toXML x) x) else x: x; + traceXMLValMarked = str: if builtins ? trace then x: (builtins.trace ( str + builtins.toXML x) x) else x: x; + + # this can help debug your code as well - designed to not produce thousands of lines + traceShowVal = x : trace (showVal x) x; + traceShowValMarked = str: x: trace (str + showVal x) x; + attrNamesToStr = a : lib.concatStringsSep "; " (map (x : "${x}=") (attrNames a)); + showVal = x : + if isAttrs x then + if x ? outPath then "x is a derivation, name ${if x ? name then x.name else "<no name>"}, { ${attrNamesToStr x} }" + else "x is attr set { ${attrNamesToStr x} }" + else if isFunction x then "x is a function" + else if x == [] then "x is an empty list" + else if isList x then "x is a list, first element is: ${showVal (head x)}" + else if x == true then "x is boolean true" + else if x == false then "x is boolean false" + else if x == null then "x is null" + else if isInt x then "x is an integer `${toString x}'" + else if isString x then "x is a string `${substring 0 50 x}...'" + else "x is probably a path `${substring 0 50 (toString x)}...'"; + + # trace the arguments passed to function and its result + # maybe rewrite these functions in a traceCallXml like style. Then one function is enough + traceCall = n : f : a : let t = n2 : x : traceShowValMarked "${n} ${n2}:" x; in t "result" (f (t "arg 1" a)); + traceCall2 = n : f : a : b : let t = n2 : x : traceShowValMarked "${n} ${n2}:" x; in t "result" (f (t "arg 1" a) (t "arg 2" b)); + traceCall3 = n : f : a : b : c : let t = n2 : x : traceShowValMarked "${n} ${n2}:" x; in t "result" (f (t "arg 1" a) (t "arg 2" b) (t "arg 3" c)); + + traceValIfNot = c: x: + if c x then true else trace (showVal x) false; + + /* Evaluate a set of tests. A test is an attribute set {expr, + expected}, denoting an expression and its expected result. The + result is a list of failed tests, each represented as {name, + expected, actual}, denoting the attribute name of the failing + test and its expected and actual results. Used for regression + testing of the functions in lib; see tests.nix for an example. + Only tests having names starting with "test" are run. + Add attr { tests = ["testName"]; } to run these test only + */ + runTests = tests: lib.concatLists (lib.attrValues (lib.mapAttrs (name: test: + let testsToRun = if tests ? tests then tests.tests else []; + in if (substring 0 4 name == "test" || elem name testsToRun) + && ((testsToRun == []) || elem name tests.tests) + && (test.expr != test.expected) + + then [ { inherit name; expected = test.expected; result = test.expr; } ] + else [] ) tests)); + + # create a test assuming that list elements are true + # usage: { testX = allTrue [ true ]; } + testAllTrue = expr : { inherit expr; expected = map (x: true) expr; }; + + # evaluate everything once so that errors will occur earlier + # hacky: traverse attrs by adding a dummy + # ignores functions (should this behavior change?) See strictf + # + # Note: This should be a primop! Something like seq of haskell would be nice to + # have as well. It's used fore debugging only anyway + strict = x : + let + traverse = x : + if isString x then true + else if isAttrs x then + if x ? outPath then true + else all id (mapAttrsFlatten (n: traverse) x) + else if isList x then + all id (map traverse x) + else if isBool x then true + else if isFunction x then true + else if isInt x then true + else if x == null then true + else true; # a (store) path? + in if traverse x then x else throw "else never reached"; + + # example: (traceCallXml "myfun" id 3) will output something like + # calling myfun arg 1: 3 result: 3 + # this forces deep evaluation of all arguments and the result! + # note: if result doesn't evaluate you'll get no trace at all (FIXME) + # args should be printed in any case + traceCallXml = a: + if !isInt a then + traceCallXml 1 "calling ${a}\n" + else + let nr = a; + in (str: expr: + if isFunction expr then + (arg: + traceCallXml (builtins.add 1 nr) "${str}\n arg ${builtins.toString nr} is \n ${builtins.toXML (strict arg)}" (expr arg) + ) + else + let r = strict expr; + in builtins.trace "${str}\n result:\n${builtins.toXML r}" r + ); +} |