| [Coke] | that blin run ended a new way: | 00:43 | |
| ⏳ 1954 out of 2508 modules processed (77.91%) | |||
| Finished with result: signal | |||
| Main processes terminated with: code=killed, status=9/KILL | |||
| I'll maybe rerun it tomorrow. | 00:44 | ||
| ugexe | did it output any results? i suspect out of all the optimizations i added there will be at least a few regressions | 00:49 | |
| japhb | [Coke]: That's a very interesting bug. Not least that Rakudo probably should apply the constant-time Range.sum optimization to [+] Range when + is the Core one, and then notice that one end is infinite and bail immediately. | 00:53 | |
| [Coke] | do you mean rakupp? | 01:01 | |
| ugexe: pushed. | 01:07 | ||
| japhb | The first sentence was about rakupp. The other was about Rakudo because you demonstrated it times out when it deals with that construction. | 01:09 | |
| So ... they both have bugs, just wildly different ones | |||
| timo | m: say (1..Inf).sum | 01:12 | |
| camelia | Inf | ||
| Geth | rakudo/main: 0e3de8847b | (Nick Logan)++ (committed using GitHub Web editor) | 5 files RakuAST: lower a for loop over a literal integer sequence (#6555) Previously a sunk for statement over an integer sequence built the Seq and pulled its iterator on every execution, while the range form compiled to a native counting loop. A sequence whose two bounds are compile time integers takes the counting loop as well. The direction is part of a sequence's meaning, so the bounds pick ... (6 more lines) |
01:48 | |
| rakudo/main: 22541d6e63 | (Nick Logan)++ | 4 files RakuAST: bind a <-> block's native parameter as a reference Previously only an is rw trait marked a parameter target rw, while the rw default of a <-> block starter stopped at the meta-object. A native parameter of a <-> block was declared as a plain lexical, the caller's native reference was dereferenced when bound to it, and an assignment inside the block never reached the caller, so ... (12 more lines) |
|||
| rakudo/main: fef921caa3 | (Nick Logan)++ | 4 files RakuAST: accept parameter traits in a variable declarator signature Previously a non-binding declarator signature such as my ($a is rw) moved every parameter trait onto the variable declaration it carries, so the rw family of parameter traits reached the variable trait dispatch and compilation died with an unknown trait error. The legacy frontend accepts these as parameter traits, and a plain ... (5 more lines) |
|||
| rakudo/main: b11a8fa230 | (Nick Logan)++ (committed using GitHub Web editor) | 5 files Merge pull request #6556 from ugexe/ugexe/rakuast-pointy-rw-native-param RakuAST: bind a <-> block's native parameter as a reference |
|||
| rakudo: ugexe++ created pull request #6557: RakuAST: close per call codegen gaps against the legacy frontend |
02:55 | ||
|
04:15
kurahaupo left
|
|||
| Geth | rakudo/main: 5 commits pushed by (Nick Logan)++
|
04:59 | |
| rakudo: ugexe++ created pull request #6558: RakuAST: lower term parameters and skip needless invocant probing |
05:02 | ||
|
07:18
lizmat joined
|
|||
| lizmat | . | 07:18 | |
| Geth | rakudo/main: d72f7dfb48 | (Elizabeth Mattijsen)++ | 8 files Mark a large number of internals as implementation-detail udo These are not documented, and not tested classes and roles. Introspection should not show these by default. Found while working on updating Rakudo's type graph. |
07:40 | |
|
07:48
lizmat left
08:27
ilogger2_ left,
ilogger2 joined
10:13
finanalyst joined
11:38
kurahaupo joined
|
|||
| ab5tract started pointing Claude at a Kotlin conversion of the JVM backend | 13:14 | ||
| just for the readability of type casting in Kotlin vs Java (in the latter it can start to look like lisp at times) | 13:15 | ||
| ... has been worth it | 13:18 | ||
| little quality of life things like not using a Makefile for managing a Java project also included | 13:21 | ||
| Geth | rakudo/main: f6c72bb69a | (Nick Logan)++ | 4 files RakuAST: lower sigilless term parameters to frame locals Previously a sigilless term parameter stayed a by-name lexical, so its binding and every use paid a lexical operation, while a sigiled object parameter already lowered to a local. The target is its own declaration and every use resolves straight to it, so it registers with the lowering walk and carries the lowered local protocol. The ... (8 more lines) |
14:07 | |
| rakudo/main: 2031a07f63 | (Nick Logan)++ | 2 files RakuAST: skip the runtime coercion probe on implicit invocants Previously a generic parameter fetched its instantiated parameter object on every call and probed its type flags for a coercion. A role method makes its implicit invocant generic, so every role method invocation paid the probe: a code object fetch, a signature walk, two attribute reads, and a flag test, and the extra size kept small role methods from inlining. An implicit invocant's type is compiler-written and never a coercion, so it needs no runtime probing. An explicit invocant keeps the probe, since it can carry any type. |
|||
| rakudo: ugexe++ created pull request #6559: RakuAST: generate the POPULATE method from the build plan |
18:18 | ||
|
20:50
finanalyst left
|
|||
| Geth | rakudo: ugexe++ created pull request #6560: RakuAST: fix native copy parameters with definite types |
21:36 | |
| rakudo: ugexe++ created pull request #6561: RakuAST: keep containers in begin-time evaluated values |
21:43 | ||
| rakudo: ugexe++ created pull request #6562: RakuAST: match legacy compile time errors for native literal misuse |
21:49 | ||
| rakudo/main: 4 commits pushed by (Nick Logan)++ | 23:59 | ||