a5d42e58d7f8ce1e985cf7422145619122ca5d25 braney Sat Sep 19 18:16:16 2026 -0700 mallocTopPadTester: check the heap step setting still reaches the library cfgSetMallocTopPad() asks glibc to grow the heap in steps of mallocTopPad bytes instead of its 128 kB default, which saves a heavy hgTracks render more than a hundred thousand system calls that draw nothing. The failure to catch is backsliding: the mallopt call is dropped in a later edit, or the setting is renamed, or it stops being read before the first allocation. Every page still draws correctly and the only symptom is that renders are slower than they used to be, which nobody attributes to this. So the test measures the step, not the clock. A timing test here would be red on a busy hgwdev, green on an idle one, and deleted within a month. sbrk(0) says where the heap ends and the distance it jumps when malloc extends it is what M_TOP_PAD sets, so the answer is the same on a loaded machine as an idle one. It is read in buckets, since glibc may round and may change what it rounds to; what must not change is the order of magnitude between a configured step and the default. Allocations stay under the mmap threshold, or the break never moves and the test measures nothing while appearing to pass. Two runs, with the setting and without, because the answer is the difference between them: a single run would pass on a library whose own default happened to be large. Watched to fail and then pass: leaving the knob read but never applied gives the default step under both confs, which is the one line the diff shows. Recorded as sandbox-ab in utils/testRegistry. refs #38225, refs #38391 diff --git src/utils/testRegistry/registry.tsv src/utils/testRegistry/registry.tsv index 89163a4aac3..9ad333d6854 100644 --- src/utils/testRegistry/registry.tsv +++ src/utils/testRegistry/registry.tsv @@ -1,89 +1,89 @@ # registry.tsv - which unit test defends which Redmine ticket. refs #38391 # # A bug ticket has no way to say whether a test now defends its fix, and a test has no way # to say which bug it came from. This table answers both, and says which tickets are still # waiting for a test. It is hand written: nothing generates it, so adding the row is part # of writing the test. # # Unit tests only. The browser-page regression tests are the docent suite, refs #38252, # where the script is already named for its ticket. A ticket whose fix only changes what a # page says is that suite's job and is not in here at all. # # Seven tab separated columns, sorted by ticket then by test: # # ticket the Redmine number, digits only # release the version the fix ships in, digits only, from the ticket's Target version. # The table starts at v504. A fix on master with no target version yet takes # "-" until the ticket says. # test the file to open, as a path from kent/src, or "-" for a ticket that needs a # test and does not have one. A suite whose cases are make targets adds # ::target, e.g. tests/makefile::relPath # docent the browser test that watches the same ticket, from the docent suite in # hg/utils/docent/tests/regress, or "-" when there is none. Not a second copy # of that suite: it is here so "nothing is watching this ticket at all" is a # question the tool can answer, which is the question worth acting on. A # docent script does not make a unit test unnecessary where the why is # invisible; rm38309 cannot see a read past the end of an array, it asserts # something next to it. # why why this ticket needs a unit test rather than a browser test: # invisible the fix changes nothing on screen. No browser test can see # it, so a unit test is the only test there can be. # perf the fix is about speed or memory. The test has to catch # backsliding, so it measures work done -- queries, passes, # allocations, bytes -- and never wall-clock seconds. # library the fix is in library code that a browser reaches only # through a page, where much else can go wrong first. # evidence what has been seen, weakest first: unrecorded, assertion-only, sandbox-ab, # release-ab, caught-regression. "-" on a row with no test. The vocabulary is # proof.js's from the docent suite, minus the two levels that need a browser, # so the two tables can be read on one scale. Every row starts at unrecorded # and earns its way up by measurement, not by argument. # note what the test holds down. On a row with no test, what it would have to hold # down, which is the first thing the person who writes it needs. # # One ticket can have several rows and one test can defend several tickets; both happen # here already. A ticket cannot be both covered and waiting. # # `testRegistry check` reads every row and fails when one has rotted, so a test cannot be # renamed or deleted without coming here. # #ticket release test docent why evidence note 10138 504 - rm10138.docent.yaml invisible - needs one: the session data directory hash went from 8 to 10 hex characters 20824 504 hg/utils/netToBigNet/tests/makefile::simpleTest - library unrecorded a net converted to bigNet and back, byte compared 27988 504 - - invisible - needs one: a server that is not the node hg.conf names has to recognise itself 36212 504 - rm36212.docent.yaml library - needs one: an explicit itemRgb on has to beat the presence of a color setting 37263 504 lib/tests/pathSimplifyTest.c - library unrecorded dot-dot collapsing, checked against the right answer rather than against the old one 37969 504 - rm37969.docent.yaml library - needs one: a quickLifted container must not hide the tracks inside it 37984 504 lib/tests/hmacTest.c - library sandbox-ab the pending social identity is signed with hmacMd5, not a plain md5 of salt plus fields 38086 504 - - invisible - needs one: a stale cart visibility variable must not hide a new BLAT result track 38126 504 lib/tests/htmlSanitizeTest.c rm38126.docent.yaml library unrecorded the allowlist that hub and custom track description HTML is filtered through 38184 504 - rm38184.docent.yaml invisible - needs one: db= resolving to the assembly already loaded must keep the session position 38185 504 hg/hgSession/tests/backupParseTest.c rm38185.docent.yaml invisible unrecorded an empty pair in a session backup must not eat the variable in front of it 38185 504 lib/tests/cgiParseTest.c rm38185.docent.yaml invisible unrecorded an empty CGI pair must not abort the request 38198 504 - rm38198.docent.yaml library - needs one: a second lift has to update a track already in the hub -38225 504 - - perf - needs one: the malloc step size now comes from hg.conf; the test has to read the knob back and see it applied +38225 504 hg/lib/tests/mallocTopPadTester.c - perf sandbox-ab the hg.conf step size reaches the C library: the heap grows in one 16 MB jump, not the default one 38233 504 - - perf - needs one: RefSeq status is asked once per track, not once per gene; the test has to count the queries, not the seconds 38236 504 - rm38236.docent.yaml library - needs one: a quickLift chain with no aligned block in the window must not crash 38248 504 - rm38248.docent.yaml library - needs one: a deprecated versioned NP_ accession has to resolve to RefSeq Historical 38249 504 hg/lib/tests/quickLiftTester.c rm38249.docent.yaml library unrecorded the target strand of a reverse complemented protein, found in the #38349 review 38253 504 - - perf - needs one: item coverage is built from feature runs, not one counter per base; the test has to go red if a per-base pass comes back 38254 504 - - invisible - needs one: a composite subtrack's visibility has to settle before the parallel loaders start 38256 504 hg/utils/hubCheck/tests/makefile::relPath - library unrecorded a local hub given by a relative path: bigDataUrl resolved once, not twice 38260 504 hg/utils/hubCheck/tests/makefile::missingFile - library unrecorded hubCheck must say something about a bigDataUrl it cannot open 38268 504 - rm38268.docent.yaml invisible - needs one: the tightened dataVersion path check 38272 504 - rm38272.docent.yaml invisible - needs one: a GenArk quickLift source assembly must not be looked for in MySQL 38273 504 - - perf - needs one: a collection's hub file is copied when it is written, not on every session load; the test has to count the copies 38283 504 - rm38283.docent.yaml library - needs one: description page variables substituted once, and in the container 38285 504 hg/lib/tests/input/hgvs/validTerms.txt rm38285.docent.yaml library unrecorded bare codon numbers and ranges, with expected/hgvs/validTerms.txt beside it 38302 504 - rm38302.docent.yaml invisible - needs one: an activation token that is missing, empty or older than seven days is invalid 38303 504 hg/lib/tests/trashDirTester.c rm38303.docent.yaml invisible sandbox-ab a session file path spelled through a symlinked config directory, which broke 583 saved sessions 38309 504 - rm38309.docent.yaml invisible - needs one: exonFrames must not be read past the end on a transcript's last exon 38313 504 - - library - needs one: a user's own __ sessions must not be hidden from My Sessions 38317 504 - - invisible - needs one: doKnownGene must not read an uninitialised stack refLink 38318 504 hg/lib/tests/sessionDataTester.c - invisible unrecorded the returned path must be freeable through kent's own handler stack 38320 504 lib/tests/faSpeedReadTest.c - invisible unrecorded the buffer grower and its caller must agree on the size 38323 504 - - library - needs one: an api key made on one geo mirror has to work on all of them 38328 504 - - invisible - needs one: the liftOver accession list passed as an slName list 38335 504 lib/tests/cgiParseTest.c rm38335.docent.yaml invisible unrecorded a pair with no =value must not lose the variable 38340 504 hg/hgSession/tests/backupParseTest.c - invisible unrecorded the same pair, read back out of a session backup 38340 504 lib/tests/cgiCookieTest.c - invisible unrecorded the same pair in a cookie header