[GSoC] Inquiries regarding manifest format and implementation strategy

[Sri Lasya Prathipati]

Dear Gao,

Thank you for the clear direction on the format priority and the
integration path.

Based on your feedback, I am structuring my GSoC proposal to
prioritize the composefs-dump style manifest as the primary milestone.
I will design the parser to act as a "virtual source" within the
mkfs.erofs framework, ensuring the architecture is modular enough to
add support for Unix-style proto files as the second phase.

I am currently finalizing the technical timeline and the architectural
design for this virtual source logic. I look forward to sharing the
full proposal soon.

Best regards,
Sri Lasya

On Mon, Mar 16, 2026 at 7:18 AM Gao Xiang <hsiangkao at linux.alibaba.com> wrote:
>
>
>
> On 2026/3/16 03:34, Sri Lasya Prathipati wrote:
> > Hi Gao and the EROFS Community,
> >
> > As I am finalizing my GSoC 2026 proposal for the project "Support
> > generating filesystems from manifests," I have been researching the
> > various manifest formats mentioned in the project ideas.
> >
> > To ensure my proposal aligns with the community's vision for
> > erofs-utils, I have two specific questions:
> >
> > Format Priority: Should the initial focus be on implementing a parser
> > for Unix-style proto files (similar to genext2fs), or is there a
> > stronger preference for supporting composefs-dump style manifests from
> > the start?
>
> I think you could finish `composefs-dump` style first, and then
> Unix-style proto files.
>
> As written in the idea page, it should support two formats at
> least.
>
> >
> > Integration Path: In mkfs.erofs, do you envision the manifest parser
> > acting as a "virtual source" that replaces the standard directory
> > crawling logic, or should it coexist as a hybrid approach?
>
> I think it's up to you, the hybird approach won't be hard if you support
> the pure "virtual source" one.
>
> Thanks,
> Gao Xiang
>
> >
> > I believe clarifying these points will help me provide a more accurate
> > and realistic timeline in my application. I've already begun looking
> > into the mkfs source to see where the insertion point for a new input
> > frontend would be most efficient.
> >
> > Looking forward to your guidance!
> >
> > Best regards,
> > Sri Lasya prathipati
>