Вирт - русифицированный паскаль. A modern, self-hosting Object Pascal compiler built for the 2020s. Zero legacy, full ARC, and unified UTF-8.
Find a file
Graeme Geldenhuys 434058ceda feat(native): caller-side const-string ARC + unified call-arg hoist region
Port the QBE backend's shape-aware const-string argument convention to
the native x86-64 backend: const params of any kind no longer get the
callee entry-retain/exit-release pair; the caller protects string
arguments by shape (borrowed / consume / pin — see
blaise.codegen.arcshapes.TConstArgMode).  Both backends now share one
convention, which also makes native calls into the QBE-compiled RTL
consistent for const params.

The pin slots live in a unified pre-call hoist region (EmitArgHoist)
shared with open-array literal blocks and — new — the sret buffers of
record-returning call arguments.  Every call path emits the same
region + post-call epilogue (string releases, record-temp field
releases, reclaim) with the call's return registers preserved:
direct calls, method calls (both strategies), constructor calls,
interface dispatch (var flags from TInterfaceTypeDesc mark positions
to skip; shape protection is applied to every string-typed value arg
since const-ness is invisible there), proc-pointer and method-pointer
calls (TProcParamInfo const flags), inherited calls, and the three
sret call paths (dest pointer saved below the region so %rsp-relative
dest operands cannot drift).

The address-taken walkers move to the shared unit
blaise.codegen.arcshapes (QBE IR output is byte-identical); the native
backend rebuilds its borrowed-argument blocklist per function from
them, mirroring the QBE backend's FConstArgUnsafe.

Bugs found by the new tests and the natively-compiled TestRunner,
fixed here (native backend, all pre-existing):
- record-returning calls as arguments interleaved their sret buffer
  between pushed argument slots (garbage args); the hoist fixes this
  and EmitSretArgReleases is gone
- callee prologue copied record/static-array value params via memcpy
  before spilling later argument registers (memcpy clobbered them);
  spills now run in two phases
- >6-slot method calls forwarded var/out params by slot address
  instead of the held pointer (no ParamMode check)
- TStringLiteral ignored IsCharCoerce: S[I] = ',' compared the
  literal's data pointer against the byte
- EmitNarrowToType truncated pointer-shaped values (class casts) to
  32 bits
- IntByteSize defaulted nil/unmapped kinds to 4 bytes (QBE uses 'l');
  forward-class fields loaded half a pointer
- EmitCallIndirect/EmitMethodPtrCall loaded the target into %r10/%r11
  before argument evaluation (caller-saved clobber window)
- EmitCall's push/pop loop used a stale ParamType from the counting
  loop

Tests: cp.test.nativeconstarg (12 asm-level shape tests incl. the
callee-pair flip) and cp.test.e2e.constarg (21 runtime parity tests on
both backends).  Full suite 2903 OK on the working and fixpoint
binaries; FIXPOINT_OK; NATIVE_FIXPOINT_OK; natively-compiled
TestRunner green except two pre-existing generic-template round-trip
failures (bugs.txt).
2026-06-11 04:04:39 +01:00
.github Add Patreon funding configuration 2026-05-13 08:54:21 +01:00
compiler feat(native): caller-side const-string ARC + unified call-arg hoist region 2026-06-11 04:04:39 +01:00
docs refactor: rename uCodeGen/uCodeGenQBE to blaise.codegen/blaise.codegen.qbe 2026-06-10 12:12:14 +01:00
plugins chore(license): relicense from BSD-3-Clause to Apache 2.0 + Runtime Library Exception 2026-05-03 19:45:29 +01:00
runtime perf(qbe): borrowed-local const-string args + lookup/format tail wins 2026-06-10 23:53:01 +01:00
scripts feat: add native fixpoint verification script 2026-06-10 14:02:01 +01:00
stdlib perf(qbe): borrowed-local const-string args + lookup/format tail wins 2026-06-10 23:53:01 +01:00
tests test(bench): record TList<TObject> scan speedup from ARC elision 2026-05-21 17:28:44 +01:00
tools chore: adopt -SNAPSHOT version suffix, update README 2026-06-10 10:39:14 +01:00
vendor/qbe vendor: update QBE from v1.2 to v1.3 2026-06-02 16:54:51 +01:00
.gitignore chore: gitignore docs/opdf-emission-design.adoc (local FYI note) 2026-06-09 17:33:44 +01:00
LICENSE chore(license): relicense from BSD-3-Clause to Apache 2.0 + Runtime Library Exception 2026-05-03 19:45:29 +01:00
NOTICE chore(license): relicense from BSD-3-Clause to Apache 2.0 + Runtime Library Exception 2026-05-03 19:45:29 +01:00
project.xml feat(tools): add bif-coverage verifier for AST/encoder/decoder drift 2026-06-08 00:27:24 +01:00
README.adoc refactor: rename uCodeGen/uCodeGenQBE to blaise.codegen/blaise.codegen.qbe 2026-06-10 12:12:14 +01:00

= Blaise Pascal Compiler
:icons: font
:source-highlighter: rouge

**The Pascal you love, reimagined for the modern era.**

Blaise is a next-generation Object Pascal compiler built from the ground up to eliminate decades of legacy baggage. It prioritizes developer productivity, memory safety, and high-performance execution.


== ✨ The Vision

The Object Pascal ecosystem has two options: Embarcadero Delphi (proprietary,
Windows-first) and Free Pascal (open source but carrying 30 years of accumulated
complexity — five language modes, five string types, and thousands of include files).

This compiler takes a different approach:

* *One language mode.* No `{$mode}` switches; no legacy dialect support.
* *One string type.* UTF-8 reference-counted string and 0-based indexing. `RawBytes` for binary data.
* *One memory model.* Automatic reference counting applies uniformly to
  strings, classes, and interfaces. No manual/auto split between `TObject`
  and `TInterfacedObject`; `[Weak]` breaks cycles. `Free` is retained as a
  synonym for immediate release.
* *Clean interfaces.* No COM GUIDs; interface dispatch via compile-time vtable mapping.
* *Reified generics.* Monomorphization at compile time — no type erasure.
* *Modern build system.* PasBuild with `project.xml`; no makefiles.
* *First-class debugger.* OPDF is the default debug format; DWARF is not required.

See link:docs/design.adoc[docs/design.adoc] for the full architecture and
implementation plan.

The result — A modern, cross-platform Object Pascal compiler targeting native code
via two backends: https://c9x.me/compile/[QBE] (the default, production-proven
backend) and a direct native x86-64 code generator (in active development,
approaching feature parity). Single language mode, single string type,
zero-GUID interfaces, reified generics, and first-class
https://github.com/graemeg/opdebugger[OPDF] debug format support.


== 🚀 Project Status

* **Self-Hosting:** Yes. Blaise bootstraps and recompiles itself with byte-for-byte fixpoint. FPC is no longer required — the entire toolchain runs on Blaise alone.
* **Testing:** 2768 tests and growing (Test-Driven Development from day one). The test suite itself compiles under Blaise.
* **Backends:** Two code-generation backends — QBE (default, production) and a direct native x86-64 backend (in active development, approaching feature parity with QBE).


[cols="1,3,1", options="header"]
|===
| Phase | Goal | Status

| 1
| Bootstrap pipeline — Hello World on Linux x86_64 via PasBuild
| Complete ✅

| 2
| Type system — classes, records, ARC, exceptions
| Complete ✅

| 3
| Generics + zero-GUID interfaces
| Complete ✅

| 4
| OPDF debug info emission
| Complete ✅

| 5
| Self-hosting
| Complete ✅

| 6
| Language improvements + expand RTL & StdLib + bug fixing
| In-Progress

| 7
| Native backend feature parity + Windows + macOS ARM64
| In-Progress

| 8
| LSP + VS Code extension
| Planned

| 9
| Migration analyser for FPC/Delphi codebases
| Planned
|===

== What Is Dropped From Classic Pascal

[cols="1,3", options="header"]
|===
| Feature | Reason for removal

| `ShortString`, `AnsiString`, `WideString`, `UnicodeString`
| Replaced by a single UTF-8 reference-counted `string` type

| `with` statement
| Source of hard-to-diagnose symbol resolution bugs; breaks static analysis

| Old-style `object` types
| Use `record` (stack/value) or `class` (heap/reference) instead

| COM-style interface GUIDs
| Interface dispatch via compile-time vtable; GUIDs are unnecessary complexity

| Multiple language modes
| One dialect, maintained well, beats five dialects maintained poorly

| `assign`, `reset`, `rewrite`, `blockread`
| Replaced by a stream-based I/O RTL

| `TObject` vs `TInterfacedObject` split
| One unified class model under automatic reference counting; `[Weak]`
  breaks cycles
|===

== 📢 Community

The core architecture is still being finalised, so the project is not yet
accepting code contributions. Feedback on language design, syntax choices, and
the future direction of Blaise is very welcome — please use the
https://github.com/graemeg/blaise/discussions[Discussions] tab on GitHub.


== Repository Layout

This project uses PasBuild's multi-module layout. Each subdirectory with a
`project.xml` is an independent module; the root `project.xml` is the aggregator.

....
project.xml                       Root aggregator (packaging=pom)
│
├── compiler/                     The compiler binary (packaging=application)
│   ├── project.xml
│   └── src/
│       ├── main/pascal/          uLexer, uParser, uAST, blaise.codegen.qbe, blaise.codegen.native.*, ...
│       └── test/pascal/          Test suite (blaise.testing, compiled by Blaise)
│
├── runtime/                      Always-linked runtime (packaging=library)
│   ├── project.xml
│   ├── Makefile
│   └── src/
│       ├── main/asm/              Platform assembly (setjmp, atomics, UTF-8)
│       ├── main/pascal/          system.pas, blaise_str.pas, blaise_arc.pas, ...
│       └── test/pascal/          Runtime tests (punit, compiled by Blaise)
│
├── stdlib/                       Standard library — opt-in via uses clause
│   ├── project.xml
│   └── src/
│       └── main/pascal/          sysutils.pas, classes.pas, math.pas, ...
│
├── tools/
│   └── migration-analyser/       FPC/Delphi migration report tool (packaging=application)
│       ├── project.xml           depends on compiler module
│       └── src/
│           ├── main/pascal/
│           └── test/pascal/
│
├── vendor/qbe/                   Vendored QBE backend source (pinned, built from source)
└── docs/                         Design documents and specifications
....

PasBuild compiles each module to its own `target/` subdirectory. Build output is
never committed to the repository.

== Building

=== Prerequisites

* A previously released Blaise binary (see `releases/`)
* https://github.com/graemeg/pasbuild[PasBuild]
* A C compiler (`gcc` or `clang`) for building the vendored QBE backend and linking
* GNU `make` for the runtime build

NOTE: FPC is *not* required. Blaise is fully self-hosting — each release binary
compiles the next version. The bootstrap chain starts from the binary in
`releases/`.

=== Bootstrap from a release

The runtime build compiles its Pascal units (`blaise_str.pas`, `blaise_arc.pas`,
`blaise_sys.pas`) using the Blaise binary at `compiler/target/blaise`. On a
clean checkout that binary does not exist yet, so the release binary must be
passed explicitly via the `BLAISE` make variable.

[source,shell]
----
# 0. Build the vendored QBE backend (once-off — not needed again unless
#    you update vendor/qbe/)
cd vendor/qbe && make && cd ../..

# 1. Build the runtime using the release binary (BLAISE= avoids chicken-and-egg)
cd runtime && make BLAISE=../releases/v0.7.0/blaise && make install && cd ..

# 2. Compile the compiler using the latest release binary
releases/v0.7.0/blaise \
  --source compiler/src/main/pascal/Blaise.pas \
  --unit-path compiler/src/main/pascal \
  --unit-path runtime/src/main/pascal \
  --unit-path stdlib/src/main/pascal \
  --emit-ir > /tmp/blaise.ssa

# 3. Assemble and link
vendor/qbe/qbe -o /tmp/blaise.s /tmp/blaise.ssa
gcc -o compiler/target/blaise /tmp/blaise.s compiler/target/blaise_rtl.a
----

Once `compiler/target/blaise` exists, subsequent RTL rebuilds (`make && make install`)
work without the override. The QBE build in step 0 is a one-off and does not
need to be repeated when rebuilding the compiler.

=== Bootstrap a development checkout

The procedure above works while the latest release binary is new enough to
compile the current source. Between releases that ceases to hold: once a commit
teaches the parser a new feature and a later commit uses it in the
runtime/compiler, the release binary can no longer build `master` directly.

`scripts/rolling-bootstrap.sh` rebuilds the chain commit-by-commit from the last
release binary up to the checked-out revision, producing a working `-pre`
bootstrap binary. See link:scripts/BOOTSTRAP.adoc[scripts/BOOTSTRAP.adoc] for the
prerequisite (placing the release binary under `releases/`) and usage.

=== Build via PasBuild

PasBuild can drive the full compile and test cycle using a Blaise binary:

[source,shell]
----
pasbuild compile -m blaise-compiler --compiler compiler/target/blaise
pasbuild test -m blaise-compiler --compiler compiler/target/blaise
----

=== Run tests

[source,shell]
----
pasbuild test -m blaise-compiler --compiler compiler/target/blaise
----

=== Verify self-hosting fixpoint

After any compiler change, verify that the compiler reproduces itself:

[source,shell]
----
./scripts/fixpoint.sh
----

This generates stage-2 and stage-3 IR and confirms they are identical.

=== Running the compiler

Once built, the compiler binary is at `compiler/target/blaise`.

[source,shell]
----
# Compile a single-file program (QBE backend — default)
compiler/target/blaise --source Hello.pas --emit-ir > Hello.ssa
vendor/qbe/qbe -o Hello.s Hello.ssa
gcc -o Hello Hello.s compiler/target/blaise_rtl.a

# Compile via the native x86-64 backend (no QBE step)
compiler/target/blaise --source Hello.pas --backend native --output Hello

# Compile with unit search paths
compiler/target/blaise --source MyApp.pas \
  --unit-path src/units \
  --emit-ir > MyApp.ssa

# Emit QBE IR only (useful for debugging the compiler itself)
compiler/target/blaise --source Hello.pas --emit-ir
----

== Licence

Apache License v2.0 with Runtime Library Exception. See link:LICENSE[LICENSE].


---
*Built with ❤️ for the Pascal community by Graeme.*